In this article, propoxate and isopropoxate were reported for the first time as drug substitutes. In order to systematically explore the structural characteristics of the propyl derivatives of etomidate, propoxate and isopropoxate were synthesized. Liquid chromatography, nuclear magnetic resonance spectroscopy, gas chromatography-mass spectrometry, ultra performance liquid chromatography-quadrupole time-of-flight mass spectrometry, infrared spectroscopy and Raman spectroscopy were used to analyze etomidate, propoxate and isopropoxate, respectively. The structural commonalities and differences of etomidate, propoxate and isopropoxate were compared. The similarities and differences of the three substances in chromatography, mass spectrometry and spectrum were analyzed. The ultra-high pressure liquid chromatography and gas chromatography were used to compare the retention times of the three substances. In the hydrogen nuclear magnetic resonance spectrum, ethyl of etomidate, propyl of propoxate and isopropyl of isopropoxate have significant differences in the chemical shift regions δ=0-5 ppm of ¹H NMR, and δ=10-70 ppm of ¹³C NMR. In gas chromatography-mass spectrometry, three substances can be quickly distinguished through fragment ion m/z 216.1 and molecular ion peaks. In ultra performance liquid chromatography-quadrupole time-of-flight mass spectrometry, three substances can be distinguished by comparing the quasi-molecular ion peaks of the primary mass spectrum and the abundance ratio of the fragment peaks of the secondary ion mass spectrum. The proposed fragmentation pattern of the three substances in the electron ionization of gas chromatography-mass spectrometry and in the electrospray Ionization of ultra performance liquid chromatography-quadrupole time-of-flight mass spectrometry were studied. Analyzing the distinct peaks in infrared spectroscopy and Raman spectra is challenging because the characteristics of peaks of FT-IR spectra and Raman spectra are complex. However, creating a database for the pure substances of the three substances enables direct comparison of spectral libraries. These studies provide fundamental data characterization for forensic toxicology and similar fields, providing technical support for law enforcement to identify substitutes of etomidate.

Detecting abnormal behavior is crucial for maintaining public security, especially in densely populated critical areas. Traditional target detection algorithms often struggle to deliver satisfactory results under these conditions due to challenges like dense target distribution, significant scale variation, and complex backgrounds. YOLOv8 is one of the better perforing detection models effect among the object detection models. This study introduces a novel approach to improve detection accuracy by integrating advanced mechanisms into the YOLOv8 backbone network. Firstly, the coordinate attention (CA) mechanism is incorporated into the C2f module of the backbone network. This enhances the network’s focus on targets amidst complex backgrounds by emphasizing relevant features and suppressing noise. Secondly, the swin transformer model is integrated into the YOLOv8 backbone. The swin transformer facilitates greater information interaction across the feature map, effectively utilizing the background information and improving object detection accuracy under complex scenarios. The datasets used in the experiments are described, the evaluation indexes of P, R, AP and mAP are listed, and ablation experiments and comparative experiments are carried out. Experiments demonstrate the feasibility and effectiveness of these improvements. The enhanced network is compared with several mainstream networks, showing a significant improvement in average accuracy, reaching 95.1%. Compared to the basic network YOLOv8, the average precision has been improved by 2.4%, which proves the effectiveness of this method. In summary, the innovative integration of the CA mechanism and Swin Transformer model into the YOLOv8 backbone network addresses key challenges in detecting abnormal behavior in densely populated and complex environments. These enhancements lead to improved detection accuracy, making it a promising approach for public security applications.

A gas chromatography-tandem mass spectrometry (GC-MS/MS) based method was developed for the determination of etomidate and metomidate in e-cigarette liquid. The samples were extracted with methanol and separated on a HP-5MS column (30 m×0.25 mm×0.25 μm). Etomidate and metomidate were detected by GC-MS/MS under multiple reaction monitoring (MRM) mode, and quantified by an internal standard method. The mass spectrometry parameters, including parent ions, product ions, and collision energy, were optimized in this experiment. The results showed that etomidate and metomidate exhibited good linearity within the range of 0.1 to 10 μg/mL, with R² no lower than 0.999. The recovery rates of spiked matrix were between 98% and 107%. The relative standard deviations (RSDs) for intra-day and inter-day precision were less than 8.83% and 9.34%, respectively. The concentrations of etomidate and metomidate in seven e-cigarette liquid samples ranged from 4.96% to 18.48%. This method is simple, easy to operate, with high extraction recovery, good reproducibility, and high sensitivity. It can be used for the detection and analysis of etomidate and metomidate in e-cigarette liquids.

With the advancement of paperless offices, handwritten electronic signatures have become widely adopted across various sectors, including banking, government affairs, and commerce. These signatures refer to the textual impressions created by fingers or stylus pens on electronic screens, such as smartphones and digital tablets. Concurrently, dynamic attributes like writing duration, stroke sequence, and pen pressure are captured and stored electronically on computers. From an implementation perspective, handwritten electronic signatures fall within the broader category of electronic signatures. However, they differ from digital signatures in terms of implementation technology, presentation format, verification methods, and technological requirements. Given the distinct verification methodologies, handwritten electronic signatures should not be subject to the authentication framework outlined in The Electronic Signature Law. Instead, their unique handwritten characteristics should be analyzed from a morphological angle. Nevertheless, shifts in writing devices and mediums have diminished certain static characteristics, often resulting in imperfections in the identification process. This complexity is further compounded by difficulties in leveraging dynamic features for identification, challenges in distinguishing copied and pasted handwriting, and a scarcity of experienced authenticators. Consequently, during the identification process, it is crucial to: collect comparable original specimens, specify the writing devices used, gather samples from identical or similar sources as the specimens, and utilize identification expertise to identify stable traits with individual variations. By adopting a comprehensive quantitative analysis approach and corroborating with electronic data, the accuracy of identification conclusions can be significantly enhanced.

The forensic science is undergoing a paradigm shift from the traditional paradigm to the features LR paradigm and the similarity scores LR paradigm, in an era of parallel three paradigms. Given the advantages and development opportunities brought by Bayesian LR paradigm, paradigm shift has become a major trend in forensic science in the global. However, paradigm shifts have not yet been fully realized on a global scale, and the development of the forensic paradigm is imbalance in various forensic discipline and in different countries and regions. The main reasons that hinder the paradigm shift of forensic science include limitation of the technology methods of new paradigm, misconceptions of some personnel about the new paradigms, insufficient abilities related the new paradigms, and legal application issues. Except for DNA evidence identification, the application of new paradigms and paradigm shift in China are relatively lagging behind. This article proposes an implementation path for the paradigm shift of forensic science in China, including scholars and practitioners conducting scientific research on new paradigms, regulatory authorities making paradigm shift strategies and plans, forensic lab developing and confirming LR methods, developing LR verbal scales, collecting data, training examiner and taking proficiency tests, decision-makers receiving relevant education, and legislators adjusting relevant regulations.

Currently, smartphone screenshot images, as a swift and convenient tool for capturing and sharing information, have found widespread applications in daily life and professional work. However, simultaneously, the associated security and privacy concerns regarding these images have become increasingly salient. Due to the significant differences in their generation mechanisms between smartphone screenshots and captured images, this fact poses challenges for screenshot image traceability and forensics. In the current research and practical applications, technical research on tracing the sources of screenshot images remains in its nascent stage, with accurate tracing and identifying the source devices of specific images emerging as a pivotal research topic in the field of digital image forensics. This study focuses on developing a recognition methodology and system based on metadata features. By collecting screenshot samples from 52 smartphones of prevalent mobile phone brands in the market, metadata is extracted from the sample images, and a metadata feature database is subsequently established. The source device of the image is matched and identified by comparing multi-dimensional features, including model, device manufacturer, profile creator, rotation angle, file type, image width, and image height. Experimental results demonstrate that the proposed method exhibits high accuracy and reliability in identifying the origins of images for traceability purposes, thereby offering a novel technical approach for the field of digital forensics.

Dried blood spot (DBS) is a sampling technique in which a small volume of blood is collected on a specific type of filter paper. Since the 1960s, DBS has been primarily used for the newborn screening of metabolic diseases. With the development of highly sensitive analytical instruments, the dried blood spot technique allows for accurate quantification of ethanol, stimulants, drugs of abuse and heavy metal elements in biological samples. Due to its obvious advantages, the application of DBS in forensic toxicology has witnessed significant growth in the past decade, which is highlighted and presented in this article. Some challenges and suggestions of dried blood spot applications were summarized for the further research as well.

This paper aims to validate the multiplex amplification system of 9 CpG sites reported in the literature, and explore its applicability in the Chinese population. The SNaPshot multiplex amplification system was validated in terms of accuracy, analysis of the initial amount of converted DNA templates, and detection of mixed samples. A total of 236 samples of five types of body fluids including saliva, semen, blood, vaginal secretion, and menstrual blood were selected. The SNaPshot multiplex amplification system was used to detect the methylation values of 9 CpG sites. The detection threshold of CpG sites was that the methylation value is greater than 0.1. Analyze the starting amount of transformed DNA templates in this system after converting DNA using sodium bisulfite (template amount ranging from 0.5 ng to 10 ng). DNA extracted from four body fluids, including saliva, semen, blood, and vaginal secretion, were mixed in the following ratios: 1∶1, 1∶5, 1∶10, and 1∶20. Finally, the detection data set of 232 samples of the five types of body fluids was used to optimize the existing body fluid source determination method. The train set (n=162) was used to construct a random forest model, and the test set (n=70) was used to predict the body fluid type and evaluated the predictive performance of the model. Furthermore, an external data set (n=40) was added to validate the prediction model. In saliva, semen, blood, vaginal secretion samples, and menstrual blood, the body fluid type was determined directly based on the specific sites of body fluids, and the accuracy rates of body fluid identification were 100%, 98%, 98%, 94% respectively. Due to the influence of the menstrual cycle, some sites were missing, and the average accuracy of menstrual blood identification was 21%. This system could effectively detect the amount of transformed DNA from 1 ng to 10 ng. Among the mixed sample, both body fluid sources were correctly identified in all 1∶1 mixed samples. The main components could be detected in the other mixed samples (ratio 1∶5, 1∶10 and 1∶20), while there was a significant difference in the detection of secondary components. A random forest model was built from 232 samples, and the accuracy of identifying the five fluid sources in both the test and validation sets was 100%. The above results show that the multiplex amplification system has high accuracy for the identification of saliva, semen, blood and vaginal secretion, and is suitable for the identification of trace samples, mixed samples (ratio 1∶1) or main components of other ratios. Compared with direct interpretation based on body fluid specific peaks, the new random forest model can better identify menstrual blood. In summary, the multiplex amplification system for tissue identification of five types of forensic body fluids based on DNA methylation is potential for good forensic application.

Diphenidol is a non-phenothiazine anti-vertigo and anti-emetic over-the-counter drug, an analogue of trihexyphenidyl hydrochloride, and usually in the form of its hydrochloride, which was first approved in the USA in 1967. Diphenidol can improve the blood flow of cerebral vertebral artery, reduce the vertigo stimulation of vestibular nerve, inhibit the labyrinth function of inner ear, and block the vomiting center or the medullaoblongata emetic chemosensory area. Then, this drug has good anti-vertigo effect, less adverse reactions, high safety and so on, so it has been widely used in clinical practice, especially employed as an anti-emetic agent in the treatment of vomiting and vertigo associated with surgery, radiotherapy, chemotherapy, Meniere’s disease, and other labyrinthine disorders. Diphenidol is generally considered a relatively safe drug, which is inexpensive and easy to purchase. However, ingesting large doses intentionally or accidentally can cause serious toxic effects. The main symptoms of diphenidol poisoning include dry mouth, irritability, hallucinations, headache, euphoria and temporary hypotension. In severe cases, respiratory failure may result from respiratory depression, hypotensive shock or arrhythmia. In recent years, there have been a number of suicides and accidental poisonings related to diphenidol in China, including preschool children being poisoned by ingestion and adults attempting suicide. In forensic practice, diphenidol poisoning can be difficult to diagnose. Additionally, clinical symptoms and signs of diphenidol poisoning vary and are easily confused with epilepsy, tetanus rabies, hysteria or rodenticide poisoning. In addition, if there are no tablets or bottles left on the scene, a definitive diagnosis of diphenidol poisoning may be overlooked by the forensic doctor. Therefore, the physicochemical properties, pharmacological and toxicological effects, metabolic pathways and products in vivo, human pharmacokinetics, zoological experiments, analysis and detection techniques and other aspects of diphenidol were described in this paper, in order to provide some theoretical references for related cases.

This paper aims to establish a solid-phase support liquid-liquid extraction-high performance liquid chromatography/triple quadrupole tandem mass spectrometry (SLE-HPLC-MS/MS) method for the detection of scopolamine, anisodamine and atropine in blood and urine. The effects of protein precipitated method, solid phase extraction method and solid-phase supported liquid-liquid extraction method on the extraction of drugs from blood and urine samples were investigated. Scopolamine, anisodamine and atropine, in blood and urine samples were analyzed by high performance liquid chromatography triple quadrupole tandem mass spectrometry. The results showed that the recovery rate of solid-phase supported liquid-liquid extraction was the highest. The linear relationship between the concentration of scopolamine, anisodamine and atropine in blood and urine and the peak area (r>0.999 2) was good in the range of 0.1－100 ng/mL (blood) and 0.5－100 ng/mL (urine). The minimum detection limit of scopolamine, anisodamine and atropine in blood was 0.01 ng/mL, and the quantitative limit was 0.1ng/mL; The minimum detection limit of scopolamine, anisodamine and atropine in urine was 0.05 ng/mL, and the quantification limit was 0.5 ng/mL. The solid-phase supported liquid-liquid extraction liquid chromatography/tandem mass spectrometry method is characterized by simple operation, less solvent usage and high recovery rate. It is suitable in the detection of scopolamine, anisodamine and atropine in blood and urine for three types of drug poisoning cases (incidents).

Postmortem interval (PMI) refers to the interval between the discovery or examination of the body and the occurrence of death. Estimation of postmortem interval is one of the important research contents in forensic pathology，and it has always been the focus and hot spot of research work. Different techniques are used to evaluate and analyze the changes in the human body after death to estimate postmortem interval. The traditional methods of estimation of postmortem interval are based on postmortem phenomena such as algor mortis, rigor mortis, livor mortis, etc. These methods rely on the subjective experience of forensic pathologists, and the estimated time of death is a relatively wide range, and the estimated result is susceptible to subjective judgment. In recent years, postmortem computed tomography (PMCT) has become increasingly influential in the field of forensic pathology. Postmortem computed tomography is a non-invasive, rapid, and objective auxiliary means of autopsy, which can significantly improve the quality and efficiency of autopsy and can find imaging features that may not be observed in traditional autopsy. After the death of the human body, the body will undergo a series of postmortem changes according to a certain time law. A series of characteristic imaging changes in cadaver organs and tissues with the passage of time of death can be observed through the application of postmortem computed tomography, and these imaging changes can be quantitatively and objectively described, The potential ideal imaging indicators can be used to estimate postmortem interval and the correlation between these indicators and the postmortem interval can be analyzed. This paper systematically reviewed the research on the correlation between postmortem imaging features of organs and tissues (such as brain, heart, aorta, lung, liver, etc.) after postmortem computed tomography of various parts of the cadaver (head, chest, abdomen) and the estimation of postmortem interval, in order to provide new ideas for the study of the estimation of postmortem interval and further expand the application of virtual anatomy technology in the field of forensic pathology.

In recent years, Android system applications (hereinafter referred to as ‘APPs’) have become one of the primary ‘tools’ used by criminals for fraud. Criminals develop fraudulent apps and distribute their installation packages, known as Android application packages or APK files, to victims. After downloading and installing these apps, victims are deceived through their interactions within the apps. Therefore, the functional analysis of apps on Android devices has become a crucial source of for analyzing the processes of fraudulent activities and identifying the perpetrators of such crimes. With the development of protective technologies in recent years, an increasing number of fraudulent application files now employ various protective measures to prevent virtual machine executing and packet capturing, making dynamic analysis of these APPs increasingly difficult. This paper introduces common anti-packet capture techniques, including APK environment detection, packet capture detection, and certificate verification detection, and starts with reverse code analysis of APKs, dynamic packet capture analysis, and the underlying system code of Android, which explores the feasibility of bypassing dynamic detection and anti-packet capture mechanisms. The study of these methods for evidence collection provides valuable insights for the analysis of various types of fraudulent and malicious APPs.

To establish a high performance liquid chromatography-tandem hybrid triple-quadrupole linear ion trap mass-spectrometry method for qualitative and quantitative determination of etomidate in blood and hair. Deuterium cocaine was selected as the internal standard, and blood sample was extracted by acetonitrile (V/V, 1/6), then centrifuged at high speed; hair sample (about 20 mg) was ground by a ball mill, then extracted by methanol and passed through the organic membrane. Using 0.1% (V/V) formic acid aqueous solution and 0.1% (V/V) formic acid acetonitrile as mobile phase, the analyte was separated and analyzed by an ACQUITY UPLC^®C18 (2.1 mm×100 mm×1.7 μm) column. Electrospray positive multiple reaction monitoring/information-dependent acquisition / enhanced product ion scanning (MRM-IDA-EPI) and secondary library retrieval were used for analysis. The results showed that etomidate in blood had a good linear relationship in the range of 1.0 to 100.0 ng/mL (r>0.995). The linear relationship of etomidate in hair was good in the range of 0.05 to 5.0 ng/mg (r>0.995). The detection limits (S/N≥3) for blood and hair were 0.2 ng/mL and 0.002 ng/mg, and the quantitative limits (S/N≥10) were 0.5 ng/mL and 0.005 ng/mg respectively. The recoveries of 1.0, 10.0, 100.0 ng/mL in spiked blood were 97.1% to 103.4%, and 0.5, 2.5, 5.0 ng/mg in spiked hair were 84.0% to 99.8%. The relative standard deviation was less than 15%. This method can be applied to the qualitative and quantitative analysis of etomidate in blood and hair in the juridical practice.

The likelihood ratio paradigm of facial similarity score is the theory and method for interpreting the evidential significance of score finding from facial comparison. Facial similarity score likelihood ratio is the latest method of Bayesian likelihood ratio paradigm for forensic science. The likelihood ratio (LR) of facial similarity score is the ratio of the occurring probability of the facial score finding quantitatively assigned based on the probability distribution data of facial scores, under a pair of conflicting propositions that usually represent the claims of the prosecution and defense parties. The propositions typically deal with the question of whether a facial image with unknown identity collected at a crime scene comes from a suspect with known identity. The face score LR expresses the relative support direction and strength of the face score finding for the propositions of the prosecution and defense parties, providing quantitative evidence value for decision-makers to determine the disputed fact of the face source. The decision maker determines the fact of the facial source proposition based on the facial score LR opinion, or the posterior probability of the facial source proposition derived from the LR and the prior odds through Bayesian law, combined with other evidence, to exclude reasonable doubt. The likelihood ratio paradigm of facial similarity score is completely different from the traditional paradigm we are accustomed to in terms of scientific logic, opinion formation, expression, understanding, and reasoning applications. It also differs from the widely used LR paradigm of DNA feature findings, which poses new requirements and challenges for forensic examiner and decision-makers.

The means and forms of drug transmission are becoming increasingly diverse, posing significant challenges to public security investigations, laws and regulations supervision, and identification technology. In terms of identification technology, the detection of isomers, which can easily lead to misjudgments, remains particularly challenging. The comprehensive analysis of multidimensional detection methods has emerged as a trend in addressing the identification and regulation of new psychoactive substances. Currently, a notable synthetic cathinone called 2-dimethylamino- 1-[3,4-(methylenedioxy)phenyl]-1-pentanone(N,N-dimethylpentylone) warrants attention. This paper aims to develop an analytical method for identifying N,N-dimethylpentylone as a new synthetic cathinone. The processed unknown samples are analyzed using a combination of gas chromatography-mass spectrometry (GC-MS), liquid chromatography-mass spectrometry (LC-MS), nuclear magnetic resonance (NMR) including ¹H-NMR, ¹³C-NMR, H-H COSY, HSQC, HMBC, and Fourier transform infrared spectroscopy (FTIR). GC-MS analysis results show that the qualitative ion fragments are m/z 100.1, m/z 58.1, m/z 121.1, m/z 149.0, with a retention time at 11.888 minutes. LC/MS analysis results show that the parent ion is m/z 250.05, and the product ions are m/z 100.10, m/z 205.10, and m/z 135.10, with a retention time at 9.772 minutes. FTIR analysis results show that there area carbonyl absorption peak at 1 676 cm^-1, and a benzene ring vibration absorption peak at 1 506 and 1 441cm^-1. One-dimensional nuclear magnetic spectrum results show 9 hydrogen signals with different chemical shifts and 14 carbon signals with different chemical shifts in the unknown compound. Two-dimensional NMR analysis results show that the unknown compound conforms to the structure of N,N-dimethylpentylone. In summary, N,N-dimethylphenylone components are detected in the unknown sample.

Footprint features, as one of the biological features of the human body, play an important role in the field of personal identification. At present, most research on footprint recognition focuses on footprint images as experimental data, using deep learning algorithms as the foundation and relying on auxiliary algorithms to complete high-precision footprint recognition tasks. However, there is a problem with models built on footprint images. Due to the similarity of footprints of different people, as the number of samples increases, the differences between the features of footprints of different people will continue to decrease, leading to an increasing false detection rate of the model. In order to reduce the interference of similarity between footprints on model recognition ability, this paper takes dynamic footprints as the research object and proposes a dynamic footprint retrieval method based on multi-class feature fusion. The proposed method uses a spatio-temporal fusion module to integrate the spatio-temporal information of footprints, so that the footprint recognition method is not limited to the apparent information of footprints. Firstly, the convolutional neural network is used to extract the frame level features of dynamic footsteps, and then the feature fusion module calculates the complete apparent features of the fused dynamic footprints through a trainable weight matrix and frame level features. Secondly, the temporal aggregation branch of the spatio-temporal feature fusion module is used to extract long-term temporal features within frame level features, and then the long-term temporal features are fused with frame level features through orthogonal fusion calculation method to form spatio-temporal features. Finally, the visual features and spatio-temporal features are fused for dynamic footprint retrieval. A comparative experiment is conducted on a dynamic footprint dataset of 200 people with existing deep learning algorithms, and the experimental results shows that this method achieved better performance, with Rank1 and mAP being 85.39% and 55.28%, respectively.

In the monitoring of the crime scene, recording can comprehensively record all kinds of sounds at the scene, including verbal conversations, abnormal sounds and other sounds that may be related to the case. The recording evidence can play a role in restoring the incident and revealing the facts of the crime. This article takes the audio recordings from the central scene of a shooting incident as the main analysis object. It introduces the method of forensic speaker recognition into the analysis of abnormal verbal cries and air gunshots related to the case. The audio-acoustic-phonetics analysis method is used to examine the cries and gunshots. A total of 142 shouts and 2-suspected shooting sounds appeared in the on-site recording of the case. Through inspection, analysis and confirmation, the shouts were all from the victim, and the two-suspected shooting sounds were air gun shots. Afterwards, combined with the on-site investigation and post-mortem examination results, the incident process was reconstructed and the sequence of events was reproduced, and the time of the incident and the license plate of the suspect’s vehicle were accurately inferred, which provided scientific basis and important clues for finding out the nature of the case, determining the direction of investigation and locking the suspect.

In recent years, telecommunications fraud cases have become increasingly prevalent, with criminals continuously refining their fraud tactics. From initially exploiting mainstream instant messaging platforms like WeChat and QQ, perpetrators have shifted to luring victims into specially crafted apps. To streamline development and reduce costs, criminals embed third-party SDK interface codes into these illicit apps, with IM (Instant Messaging) services being a common type. In these novel fraud scenarios, conducting both dynamic and static analyses of the chat-focused apk files involved can yield valuable forensic leads regarding IM services and access databases containing crucial information for investigation and prosecution. This article, taking the Android system as an example, meticulously outlines the process of mining forensic clues from instant messaging apps and subsequent data analysis, encompassing technical principles, analysis and processing steps, and case applications. It emphasizes techniques such as extracting key values through apk static analysis, validating those keys via dynamic packet capture, and utilizing SQL queries to sift through and analyze chat logs, thereby offering a professional methodological reference for evidence gathering in related cases.

In order to further improve the accuracy of speaker recognition and avoid the complicated process of manual feature extraction required by traditional speaker recognition methods, this paper proposes an end-to-end speaker recognition method based on CBAM attention mechanism and deep neural network. CBAM, a lightweight general module, is introduced into the deep neural network structure and seamlessly integrated into the network architecture. After it is added into the first layer of deep neural network convolution in this paper, the features of speech signals first pass through the CBAM channel attention module to strengthen the model’s attention to the channel dimension of speech features. Then CBAM spatial attention module is used to improve the model’s attention to the spatial dimension of speech features, further improve the model’s sensitivity to important feature information, and use the end-to-end loss function to train the whole model as a whole. At the same time, an embedded court speaker recognition method based on generalized end-to-end loss function training is proposed, and the likelihood ratio is obtained by using the embedded cosine similarity score trained by the improved network model, so as to intuitively and accurately judge whether it is the same speaker, thus providing intuitive and powerful evidence for the court. Finally, taking deep neural network BILSTM and GRU as examples, the mainstream data set CN-Celeb was used to train the model to ensure that the model can achieve better combat effects in a complex and rich voice environment. Zhaishell, a subset of Zhvoice, and the audio of actual combat cases collected by ourselves were used for combat test, to ensure that the model in this paper has a good recognition effect for both Mandarin and dialect. The results show that the method proposed in this paper can effectively improve the recognition accuracy, quickly construct the model and improve the generalization ability.

Determining the sequence of intersecting lines between laser printing and stamped impression is one of the key contents of questioned document examination. Fluorescence microscope is considered to be one of the most effective ways to examine the sequence of intersecting seal and toner lines of questioned documents. However, due to various factors, determining the sequence of crossing lines has always been a challenge to forensic document examiners. Considering the influence of types of factors such as toner morphology, stamped impression fluorescence, and paper, a systematic analysis of laser printing handwriting and print timing issues was conducted using the ZMSX-05 vermilion ink timing instrument to excite fluorescence. The fluorescence phenomenon on the surface of toner was observed under two different timing conditions (“printing before stamping” and “stamping before printing”). Experiments have shown that the morphology of ink powder plays an important role in determining the sequence of intersecting lines. When the toner powder is compact and piled up on the surface of the paper and the seal ink has strong fluorescence, the sequence of the intersecting lines can be determined quickly and accurately; however, when the toner powder is non-compact and penetrates into the paper fibers, the seal ink has weak fluorescence, making it relatively difficult to determine the sequence of the intersecting lines. However, by comparing the transmitted light and fluorescence test images, it is possible to accurately determine the sequence between laser printing and stamp impression under most conditions, especially for compact and non-compact toner examination, for which 100% accuracy rate and 90% detection rate were achieved in the blind test, but some examiners made errors in determining the sequence under the interference of paper fibres, which indicates that the conclusions drawn by examiners with different professional abilities are different. This also reminds us that in actual identification, examiners should understand the principle of the fluorescence method, distinguish the effect of each element on the sequence of intersecting lines, and scientifically apply examination methods in order to make correct identification opinions. In addition, time should be another important element in determining the sequence of intersecting lines, but the time interval between laser printing and stamping, as well as the changes that occur over time after the formation of the two time sequences, require long-term detection. For example, how the fluorescence of the printed text on the surface of the toner powder changes over time, etc. This should be an important line of research for the future, so that experimental data can provide reference and clarification for document examiners.

Rapid non-destructive testing of the formation time of blood stains at different dilution ratios under common temperature and humidity conditions was developed using UV visible reflectance spectroscopy. Dilute fresh blood with anticoagulant by 2, 4, 8, 16, and 32 times with tap water. Take 20 μL diluted blood and drop it in the center of a white cotton cloth. After drying, use the reflection accessory of a UV visible spectrophotometer (UV-2450 type) to collect the reflection spectrum, and Excel was used to do regression analysis of trough values and formation time of blood stains at different concentrations. As time increased, there was a significant blue shift in the reflected wave troughs of normal and diluted blood stains near 577.0 nm and 540.0 nm. The values of R² of the regression equation were all greater than 0.800 0, and there was a good linear relationship between the trough blue shift and the formation time of blood stain. The average speed of the trough blue shift increased with the increase of dilution ratio. Conclusion: The blue shift in the reflected wave valleys at 577.0 nm and 540.0 nm can serve as a basis for determining the formation time of normal blood stains, and combined with visual observation, it can also serve as a reference for determining the formation time of diluted blood stains.

In recent years, Chinese patent medicines and health care products have become more popular, but some businessmen have added illegal additions in them for profit. At present, there are many kinds of illegal additions. The objects of this study mainly include seven categories: medicines for regulating blood pressure, medicines for blood sugar regulation, medicines for blood lipid regulation and so on. These drugs may lead to a series of problems in cardiovascular, hepatic/renal function, which can be life-threatening in serious cases. In order to protect public health and rights, China has issued many standards and relevant guidelines. The main detection techniques are: immunoassay, spectroscopy, chromatography and the combination of various technologies and real-time direct analysis mass spectrometry. This paper summarized the categories of illegal additions in recent years, the commonly used sample pretreatment technology and the corresponding detection technology, and briefly introduced the characteristics of each technology. Finally, an outlook is given about the management of new illicitly added chemicals, the direction of development on related detection technologies, and the measures to deal with new illicitly added chemicals.

This study explored the utilization of low-altitude unmanned aerial vehicle (UAV) imagery combined with ground-based handheld photography to create a multi-scale, variable-resolution 3D realistic model of high falling incident scenes. Initially, a low-altitude UAV is employed to capture imagery data of the orientation, general overview, and fall space of the high falling incident site. Subsequently, a digital single-lens reflex (DSLR) camera is utilized to obtain detailed image data of key areas such as the starting and ending points of the fall. In the SMART 3D reconstruction software, feature points are marked and matched between the UAV imagery and the handheld camera images, aiming to rigidly correlate the aerial and ground data, thereby fusing these diverse sources to create a refined 3D model. The results demonstrate that the 3D model, constructed through the fusion of the two data sources, offers a multi-scale, comprehensive representation of the high falling incident scene. It enables users to observe both the overall orientation and layout from a distance, as well as to scrutinize key areas and detailed features up close. This technical methodology provides a direct and authentic representation of the location, texture, size, and other physical attributes of trace evidence at the scene. It offers a novel means for digitally documenting and preserving the high falling incident scene, which can then aid in reconstructing the fall sequence and enable accurate determination of the case's nature.

The identification of the characteristics of shooting bullet trace image is the main content of gunshot trace inspection, and also one of the challenges. This article introduces an advanced automatic annotation method for shooting bullet trace features based on the High-resolution networks (HRNet) framework, which can achieve automatic labeling of the land-engraved trace area, groove-engraved trace area, and slippage trace area. A database of 5 985 images containing seven different sizes of shooting bullet traces extracted by BalScan (3D trace image scanning system) was constructed and divided into training, validation, and testing datasets at a ratio of 7:1.5:1.5. The training dataset was manually annotated to identify the land-engraved trace area, groove-engraved trace area, and slippage trace area, which were used to train the high-resolution network model. Then, the unlabeled testing dataset was input into the trained model for automatic annotation of the feature areas. Finally, the annotation results were manually reviewed and the accuracy was recorded. The results showed that the proposed method achieved an average accuracy of 94.1% in the automatic annotation task, demonstrating its effectiveness. This annotation algorithm for shooting bullet trace images without manual annotation can significantly reduce the workload of inspectors and provide a feasible new approach to improve the efficiency of firearm trace inspection.

The aim of this study is to provide guidance and reference for further research in related fields by analyzing in-depth the current research status, emerging themes and future development trend of forensic entomology. In order to achieve this goal, the authors systematically searched the literature related to forensic entomology from CNKI (China National Knowledge Infrastructure), Wanfang database, and Pubmed database during the period from 2003 to 2023. Using CiteSpace 6.2.R4 Advanced software, the authors visualized and analyzed these literatures in terms of annual publication trends, authors, countries, institutions, and keywords. After rigorous screening, a total of 499 domestic and 2 137 foreign literatures, totaling 2 636 articles, were included in this study. The analysis results showed that although China has made some progress in the field of forensic entomology, there is still much room for improvement in international cooperation and exchange. Meanwhile, the inference of the time of death (PMI) of corpses is a research hotspot of common concern at home and abroad, but there are differences in research focus among countries. We conclude that China must strengthen empirical research and field investigation, focus on the combination of theory and practice, improve the application value and influence of research, and enhance communication and cooperation between scholars internationally to jointly promote the development of forensic entomology.

Human body odor arises from the secretion of various glands on the skin's surface, which, when acted upon by microorganisms, evaporate to produce a distinct scent. This odor contains valuable biological information, with certain compounds exhibiting strong stability and individual specificity, serving as “odor fingerprinting” that can distinguish between different populations. Machine learning is an important method for human odor research, which can not only explore the characteristic components of odor in different populations, but also investigate the differences between different individuals. This paper discusses the application of “odor fingerprinting” in individual identification and feature characterization, drawing upon recent literature. It outlines the data processing procedures involved in human odor analysis, highlights the challenges encountered, and explores current research trends. Finally, the application trends of the recognition of human odor are discussed in order to provide reference for odor recognition research.

Injuries caused by animals are relatively common in forensic examinations, but those resulting from bird pecking are less frequent. In particular, eyeball injuries caused by bird pecking are even rarer. Such injuries can easily be mistaken for human-inflicted injuries, leading to misunderstandings among the deceased’s family members and even forensic personnel. This can cause the case to reach an impasse or even trigger petitions. This article presents a case of ocular perforating defects caused by bird pecking, aiming to introduce the characteristics of animal injuries and the specific features of injuries caused by bird pecking, as well as how to analyze and evaluate such injuries, thereby providing a basis for determining the nature of the case.

Gun-related cases pose extreme societal harm and urgently demand efficient and precise detection methods. This study aims to integrate the reflectance transformation imaging (RTI) technique with deep learning to apply it to the field of gun and bullet recognition. In the experiment, a total of 1 500 samples of fired cartridge cases were selected from five QSZ92 9 mm pistols. Detailed images of the markings on the base of the cartridge cases were captured using the DTV3.1 intelligent imaging system to obtain their normal maps. Thereafter, the pre-trained ResNet-50 network extracted features from the normal maps and underwent classification training. The model’s performance was evaluated by outputting AUC values, accuracy on the test set, and a confusion matrix. The experimental results reveal a total AUC value of 0.98 across the five guns, with gun No. 2 achieving the highest accuracy of 97.66% and gun No. 1 the lowest at 93.75%. This study demonstrates that the automatic recognition method of cartridge case marks based on RTI technology and deep learning yields significant results, offering valuable reference for the identification of other traces in the field of trace inspection.

With the wide application of surveillance systems, there is an increasing concern about public safety and security issues. Among them, the rapid detection and recognition of fighting behavior is very important for maintaining social order and security. However, traditional monitoring systems often face many challenges when dealing with large-scale video streams, including high computational complexity and resource-limited environments. In order to cope with these challenges, this paper proposes an improved fighting behavior detection model based on YOLOv5s, which reduces the number of parameters of the model and the computational complexity, so that the model can operate more efficiently in the resource-limited environment and detect various fighting behaviors more accurately. First of all, the open source interactive markup tool Labelimg was used to annotate the data set and train the network model with a large amount of data. Secondly, considering the need for rapid and accurate solutions in public security practice, lightweight network MobileNetv3 is used as the backbone network by comparing various convolutional structures to replace the original backbone network of YOLOv5s model, so as to reduce the number of parameters and calculation amount of the model and improve the model detection accuracy. By setting ablation experiments, the improved model is compared with other models and the original model. The experimental results show that compared with the original network, the detection accuracy of the improved model is increased from 92% to 94.4%, the computational load is reduced from the original 15.8 G to 3.1 G, and the detection speed of the algorithm can reach 0.153 s at the fastest, meeting the real-time requirements. And the detection accuracy is the highest among the three models. This model is suitable for public security practical application scenarios with high precision and limited memory and computing power.

Forensic SNP genealogy can infer distant kinship based on SNP chip data. In order to clarify the ability of genealogical inference technology based on SNP chip data to detect trace DNA in forensic field, in this study Illumina CGA microarray was used to detect the samples. The samples were preliminarily evaluated based on the DNA input, detection rate, sample heterozygosity and other indicators, then the IBS and IBD algorithms were used for forensic SNP genealogy. The classification consistency was compared with the reference samples, and the factors affecting the prediction accuracy were analyzed, the detection ability of the technology system for different input amounts of DNA were determined, and finally the accurate SNP typing data in the low-quality data were screened based on the signal ratio and other indicators, so as to improve the use value of the trace DNA detection data. The results show that when the DNA input was higher than 1.95 ng, the IBS algorithm had an average confidence interval accuracy of 94.33% and 91.96% for IBD, and when the input was 488－781 pg, the IBS algorithm had an accuracy of 23.11% for the average confidence interval for 1－5 kinship, while the IBD algorithm reached 30.13%. When the DNA input is less than 488 pg, both the IBS and IBD algorithms are unable to make genealogical inferences. Allele insertion is a major factor affecting the accuracy of pedigree inference, and when the homozygous error reaches 22.5%, the sample cannot be used for pedigree inference. By screening the signal ratio, the heterozygous SNP loci with a signal ratio greater than 1.5 can be removed, which can improve the genealogical inference ability of low-input samples. Based on the real family data of Illumina CGA chips, this study analyzed the influence of sample input on the accuracy of genealogical inference, and optimized the SNP data by signal ratio, so as to improve the application value of low-input samples in genealogical inference.

For the first time, a general ¹H-quantitative nuclear magnetic resonance spectrometry (¹H-qNMR) method for the simultaneous quantification of heroin hydrochloride, cocaine, cocaine hydrochloride, and 7 adulterants was established after the discussion of some key quantitative parameters. According to the solubility and stability of heroin hydrochloride, cocaine, and cocaine hydrochloride in different deuterated solvents, and the spectra comparison of seized samples and reference materials, dimethylsulfoxide-d6 was selected as the solvent, and 1,3,5-trimethoxybenzene was selected as the internal standard. The signals at δ_H 6.09 ppm for internal standard 1,3,5-trimethoxybenzene; at δ_H 6.87, 6.71, 2.24 ppm for heroin hydrochloride; at δ_H 7.90, 7.64, 5.12, 3.61 ppm for cocaine; and at δ_H 7.88, 5.49, 2.79 ppm for cocaine hydrochloride were selected as quantitative peaks．The established ¹H-qNMR method was applied for the quantification of 168 seized samples, the content of heroin and cocaine were basically consistent with the results of High-performance liquid chromatography. This method is simple, fast, accurate, does not require standard samples, and shows good tolerance and high reproducibility. It can provide new ideas for the quantification and profiling analysis of seized drug samples.

GC-MS and GC-MS/MS analytical methods were developed for the simultaneous detection of three 4(3H)-quinazolinone abused substances, including methaqualone, etaqualone, and 2-methoxyqualone, and applied to the analysis of suspected drugs and the hair of abusers, respectively. The standard solutions of methaqualone, etaqualone, and 2-methoxyqualone were analyzed by using the selected ion monitoring (SIM) and multiple reaction monitoring (MRM) modes, respectively, to determine the characteristic fragment ions and characteristic ion pairs of the three substances. The so established two methods were used to test the suspected drugs and hair samples in the actual cases. The established GC-MS analytical method detected components of methaqualone, etaqualone, and 2-methoxyqualone in the suspected drugs at levels of 11 μg/mg, 4.3 μg/mg, and 2.5 μg/mg, respectively. The established GC-MS/MS analytical method detected methaqualone, etaqualone, and 2-methoxyqualone in the hair at levels of 148 ng/mg, 565 pg/mg, and 58.2 ng/mg, respectively. The detection methods of GC-MS and GC-MS/MS established in this study are simple, rapid, and accurate, which can provide a reliable scientific basis for the detection of methaqualone, etaqualone, and 2-methoxyqualone in the suspected drugs and the hair of abusers.

This study uses the Web of Science Core Collection as its search dataset, employing the visualization tools VOSviewer 1.6.20 and CiteSpace 6.2R6 to analyze 509 publications related to the statistical quantification of trace evidence, spanning 57 countries, 976 institutions, and 267 journals. The study examines key literature nodes from four perspectives: publication volume, publication outlets, keyword co-occurrence, and keyword clustering, providing researchers with a comprehensive and intuitive understanding of the research trends and emerging hotspots in the field. The findings reveal that over the past decades, the volume of research on statistical methods for trace evidence has shown fluctuating growth. European countries have shown significant collaboration on this topic, forming a closely-knit regional cooperation network, with the Netherlands Forensics Institute being the most prolific institution. Fingerprints are a crucial subject of statistical quantification of trace evidence, with statistical and quantitative methods primarily focusing on a series of methods based on Bayes’ theorem, such as likelihood ratios and Bayesian networks. A trend in research hotspots is observed, transitioning from clusters of subjective quantification methods (such as subjective likelihood ratios) to objective ones (such as feature-based and score-based likelihood ratios). Current challenges in the statistical quantification of trace evidence include difficulties in interpreting high-dimensional data, model error rates, and model parameter estimation. The study suggests improvements such as establishing quality assessment metrics for high-dimensional evidence, developing models with dynamically adjustable error tolerance, employing multiple validation and evaluation strategies, and fostering expert consensus on new paradigms in trace evidence.

The anonymity and decentralization of Bitcoin make it a significant medium for illicit transactions, posing challenges for traditional detection methods in handling complex transaction network structures. This study proposes a graph neural network model based on a pre-trained Conditional Variational Autoencoder (CVAE) to enhance the efficiency and accuracy of Bitcoin illicit transaction detection. The model generates K−1 feature vectors through the CVAE, which have the same number as the input features, and then combines these generated K−1 feature vectors with the original feature vector to ultimately form K feature vectors. Each feature vector undergoes multi-channel aggregation and max pooling, resulting in multiple feature vectors. These vectors are subsequently processed through linear layers and layer normalization, followed by another round of max pooling to obtain a global feature vector. Finally, the feature vectors are further processed through graph convolutional layers and linear layers to generate the final classification result. The model integrates input feature vectors at the output layer through a skip mechanism. Experimental results demonstrate that this model performs excellently in Bitcoin illicit transaction detection, significantly improving detection accuracy and robustness.

The handwriting characteristics of machine imitative signature in terms of writing style, layout, writing method are almost completely consistent with the imitated signature. A general analysis of the differences between machine imitative signatures and handwritten signatures in terms of various handwriting features makes little sense. On the contrary, there are more detailed differences on some handwriting characteristic like writing strength, writing speed and so on. In this experiment, the characteristics of stroke mark are taken as the research object, and the samples of human handwritten signatures are collected and the corresponding machine imitative signatures are made. The similarities and differences between them in stroke marks are observed and summarized by microscope. The results show that there are morphological differences between machine imitative signature and handwritten signature in ink dots, indentations, stroke width and thickness and so on. Machine-writing has strong repeatability and similarity in these features, and the same features always appear in other handwriting. Because of the different structure and function of the writing robot, the writing characteristics of the machine itself also affect the stroke characteristics of the machine imitative signature. For example, ink and white line features are not observed in machine imitative handwriting. In conclusion, the stroke marks of machine imitative signature have strong regularity and high similarity, while the stroke marks of handwritten signatures have strong randomness and rich changes. Affected by machine itself, the machine imitative signature and the human writing signature have obvious and fixed differences in stroke mark. It is a feasible handwriting inspection idea to identify machine-writting handwriting from the perspective of stroke mark.

Forensic document examination, a pivotal branch of forensic science, involves the meticulous analysis and authentication of various document forms, including handwriting, printed materials, and seal impressions. With the relentless progression of technology, the integration of deep learning methodologies has significantly accelerated the automation and intelligence in this field. Specifically, the employment of complex multi-layered neural network models within deep learning has facilitated a heightened level of document image recognition and analysis, surpassing the capabilities of conventional approaches. This technological breakthrough has not only enhanced the accuracy and efficiency of forensic document examination but also substantially reduced the influence of human error and subjectivity, thereby bolstering the credibility of results. This article provides a thorough review of the contributions by domestic and international researchers in leveraging deep learning for different aspects of document examination. It delves into the advancements in handwriting analysis, which involves the identification and comparative assessment of individual writing styles; printed document verification, which focuses on the authenticity of printed materials; and seal impression inspection, where the authenticity and source of seal marks are scrutinized. The discussion includes an overview of the foundational principles underpinning these methodologies, the specific applications of deep learning in these areas, and the cutting-edge research findings propelling the field forward.In addition to highlighting these advancements, the article also critically examines the existing obstacles and constraints in applying deep learning to forensic document examination. These include the demand for more robust and generalizable models capable of accommodating the extensive variability encountered in real-world documents, the necessity of extensive and diverse datasets to train these models, and the challenges associated with integrating deep learning tools into established forensic workflows. The article concludes by offering insights into the future directions for research and application, emphasizing the potential for deep learning to further revolutionize forensic document examination as the technology continues to mature.

In order to study the feasibility of using fluorescein to develop latent blood fingerprints based on fluorescent labeling technology and explore the optimal conditions for its development, we selected two cselected used fluoresceins (FITC fluorescein and DCFH-DA fluorescein) as the main research objects of this experiment, to design the method and procedure of developing latent blood fingerprints on non-porous surfaces with fluorescein, based on the principle of fluorescent labeling technology. On this basis, PBS buffer was used to change the pH environment of fluorescein solution, and acetone was used to dilute the two fluorescein solutions by 5 times, 10 times and 100 times, respectively, in order to explore the best developing conditions of latent blood fingerprints. The results showed that both FITC fluorescein and DCFH-DA fluorescein were effective in developing latent blood fingerprints on non-porous surfaces. The best conditions for FITC fluorescein to develop latent blood fingerprints on non-porous surfaces are: the developing time being 10 minutes, the pH environment value being 9.35, and the concentration being the original concentration; The best conditions for DCFH-DA fluorescein to develop latent blood fingerprints on non-porous surfaces are: the developing time being 20 minutes, a pH environment value being 13.01, and the concentration being the original concentration. Therefore, it is feasible for the fluorescence labeling technology to develop latent blood fingerprints on non-porous surface, but it is still necessary to further explore the preservation time of prepared fluorescein solution and the damage of DNA detection in blood fingerprints by this technology, and comprehensively analyze its practical application value.

This study established an HPLC-MS/MS method for qualitative and quantitative analysis of mifepristone and N-demethylmifepristone in blood samples. The appropriate pretreatment methods were investigated to pretreat the blood samples. HPLC was applied for separation. MRM mode was adopted to analyze. The precision, matrix effect, detection limit, recovery rate, and linear relationship of the method were all tested. The results indicated that mifepristone and N-demethylmifepristone in blood were separated well, with good chromatographic peak shape and good linear relationship in the range of 10 to 200 ng/mL. The detection limit concentration of mifepristone in blood was 0.5 ng/mL, the lower limit of quantification was 1.5ng/mL, the matrix effect was 94.6% to 98.2%, and the recovery rate was 82.7% to 92.4%; The detection limit concentration of N-demethylmifepristone was 1ng/mL, the lower limit of quantification was 2 ng/mL, the matrix effect was 91.2% to 97.3%, and the recovery rate was 81.8% to 91.7%. This method is simple, efficient, stable and easy to operate. It demonstrates satisfactory results for both qualitative and quantitative tests of mifepristone and N-demethylmifepristone, making it suitable for applying to the daily inspection and identification of related cases.

Angle grinder is a tool used for grinding metal, wood, plastic and other materials. In recent years, it frequently appears in some cases of railway cabbe theft. Because the cutting marks formed by angle grinder are similar to saw tools, it is difficult to analyze and judge the types of tools in practical cases. In the paper, through simulation experiments, angle grinder and hacksaw were used to cut copper-core cables of different specifications, and the tool marks formed at the broken ends of cables were observed by microscope, to summarize the characteristics and differences of the tool marks formed by the two. It was found that the cross-sectional shapes of the cable broken end cut by angle grinder and hacksaw were similar overall, both of which were flat, but most of them were stepped by hacksaw, while those cut by angle grinder were smooth. Angle grinder will form parallel line traces with a certain radian at the section, while hacksaw will form multiple straight lines and cross each other. At the same time, there are differences between them in starting point trace, terminal trace and chip morphology. In actual case analysis, we should make a comprehensive judgment based on the cable broken end position, cable diameter and other conditions, and emphasize the differences in traces formed by angle grinder and saw tool. The research on the characteristics and formation mechanism of tool marks formed by tool marks formed by angle grinder cutting cables can provide certain theoretical support for judging the types of tools in related cases.

In this study, we report a three-step mutation at D8S1132 locus between an alleged father (AF) and child in a paternity case. Three autosomal STR multiplex amplification kits were used for capillary electrophoresis detection. At the D8S1132 locus, the genotypes were: alleged father 17/23, parent mother 22, and child 20/22, which does not comply with Mendelian inheritance laws. Paternity was confirmed by calculating the cumulative paternity index and a three-step mutation was identified at the D8S1132 locus. Next-generation sequencing was further used to validate the experimental results and explore the source of the mutation. The results indicated that the allele 20 of the child was derived from the allele 23 of the alleged father. In addition, next-generation sequencing platform that simultaneously detected different genetic markers such as STR, SNP, and mitochondria DNA increased the cumulative parental index, which further confirmed the paternal relationship. Therefore, this case suggests that in paternity testing, when multiple mutations occur at a certain STR locus, capillary electrophoresis and next-generation sequencing can be combined for cross-validation to improve the credibility of the identification results.