Intelligent analysis system and method based on fire scene human body thermal injury traces

Abstract

The invention relates to an intelligent analysis system and method based on fire scene human body thermal injury traces, and belongs to the technical field of forensic medicine identification and artificial intelligence crossing. According to the system, a labeling database containing multiple types of thermal injury traces is constructed by collecting human body burn trace images of dead and wounded persons in a fire scene; extracting multi-modal features of the image by using a convolutional neural network; the training of the deep learning model comprises the following steps: automatically identifying and classifying thermal damage traces (including flame damage, incandescent body contact damage, gas explosion damage, combustion improver related damage, mechanical superposition damage, moored burn and the like); deducing injury reasons and processes; judging the death reason and time (distinguishing burn-to-death before birth and burn-to-death after birth); death mode analysis (normal / self-burning / other-killing); and comprehensively studying and judging the fire property (accident / misfire / fire release). According to the method, the accuracy and efficiency problems caused by the fact that traditional fire investigation depends on subjective experience are solved, and scientific and efficient technical support is provided for case detection.

Description

TECHNICAL FIELD

[0001] The present application relates to an intelligent analysis system and method based on traces of human thermal injury at a fire scene, belonging to the technical field of forensic identification and artificial intelligence. BACKGROUND

[0002] In fire-related case investigations, accurately identifying the following elements is crucial for case cracking: the cause and process of injury of the injured person; the cause and time determination of the death of the deceased (distinguishing between pre-death burning and post-death cremation); death mode analysis (natural / suicide / homicide); and comprehensive analysis of fire nature (accident / fire loss / arson).

[0003] However, traditional identification methods mainly rely on forensic experience, and have three major limitations: on the one hand, manual identification is greatly influenced by subjective factors of the identification personnel, and different identification personnel may have different interpretations of the same trace or evidence, leading to difficulty in guaranteeing the accuracy and consistency of the identification results. On the other hand, the fire scene is often complex and variable, and the injury characteristics and traces of the human body after high-temperature burning may be blurred, confused or difficult to identify, increasing the difficulty and uncertainty of manual identification. In addition, traditional identification methods are low in efficiency when dealing with a large number of complex fire evidence, and are difficult to meet the requirements of modern society for rapid cracking and efficient handling of fire-related cases.

[0004] Currently, in the field of fire-related cases, there is no system that can realize intelligent analysis of human thermal injury traces through multi-depth learning, especially lacking an integrated solution covering key identification tasks such as injury cause and process inference; death cause and time determination (distinguishing between pre-death burning and post-death cremation); death mode analysis (normal / self-immolation / homicide); and comprehensive analysis of fire nature (accident / fire loss / arson). SUMMARY

[0005] In view of the defects of the prior art, the present application provides an intelligent analysis system and method based on traces of human thermal injury at a fire scene, in particular a system and method for intelligent analysis of death and injury causes and fire nature based on traces of human thermal injury at a fire scene. The core innovation is: data collection → construction of multi-modal injury feature database → training of cascaded deep learning model → realization of trace intelligent comparison → generation of explainable analysis report. It is used to solve the accuracy and efficiency problems caused by the dependence of traditional fire investigation on subjective experience.

[0006] To solve the above problems, the technical scheme adopted by the present application is as follows:

[0007] The intelligent analysis system based on traces of human thermal injury at a fire scene is a system for intelligent analysis of death and injury causes and fire nature based on traces of human thermal injury at a fire scene, which specifically comprises the following modules in turn:

[0008] Image acquisition module: used for obtaining the digital image of the burn marks of the dead and injured in the fire scene;

[0009] Human burn mark database module: used for storing the annotated burn image dataset;

[0010] Feature extraction module: using convolutional neural network (CNN) to extract multi-modal morphological features from input images;

[0011] Intelligent research and judgment module: based on the cascade deep learning model architecture, sequentially performs:

[0012] (a) Thermal injury mark type identification and positioning: positioning the specific thermal injury type area through spatial attention mechanism;

[0013] (b) Injury process reconstruction: relying on the time series recurrent network to simulate the damage formation dynamics process;

[0014] (c) Death state discrimination: combining respiratory tract carbon deposition, heart blood CO-Hb concentration spectrum analysis and skin tissue elasticity characteristics, outputting the probability value of burn death / postmortem cremation;

[0015] (d) Death mode determination: generating the probability matrix of suicide, homicide or accident;

[0016] (e) Comprehensive determination of fire nature: integrating the output results of (a)-(d) to generate the probability evaluation of arson, accident or natural disaster.

[0017] Further, the annotation dimensions of the human burn mark database module include:

[0018] Thermal injury type label: flame thermal injury, hot body contact thermal injury, gas explosion thermal injury, combustion-supporting agent contact thermal injury, combustion-supporting agent explosion radiation thermal injury, combustion-supporting agent thermal injury after healing, mechanical superimposed injury, pre-death state burn injury;

[0019] Anatomical positioning label of injury site: referring to the 86 key areas divided by international anatomical terminology; injury time and degree correlation label: divided into I congestion period to IV carbonization period according to histological changes.

[0020] Further, the feature extraction module includes:

[0021] Texture fractal dimension of skin carbonization area;

[0022] Edge gradient distribution of blister shape;

[0023] Color space difference between wound and normal tissue;

[0024] Infrared thermal radiation characteristics of deep tissue injury.

[0025] Further, the combustion-supporting agent contact type thermal injury and the combustion-supporting agent deflagration radiation type thermal injury both include combustion-supporting agent residual form identification and deflagration radiation range modeling, which provide strong support for fire cause analysis.

[0026] Further, the death state discrimination in the intelligent analysis module is achieved by analyzing respiratory tract carbon deposition, carbon monoxide hemoglobin concentration in the heart blood, and skin tissue elasticity.

[0027] Further, the hardware deployment of the intelligent analysis system also includes:

[0028] Edge computing node: for realizing on-site image preprocessing;

[0029] Central server: for running multi-task inference model;

[0030] Forensic interaction terminal: for displaying time axis deduction animation.

[0031] An intelligent analysis method based on human thermal injury traces in fire scenes, which adopts the above-mentioned intelligent analysis system based on human thermal injury traces in fire scenes, includes the following steps:

[0032] S1: Data acquisition and standardization processing:

[0033] Collect no less than 100 cases of burn images identified by forensic experts, and label the injury type, anatomical position, and injury mechanism of each image;

[0034] Correct the distortion of the corpse posture by performing thin plate spline transformation, and segment the adhesion injury area by using U-Net network; form an anatomical positioning reference map;

[0035] S2: Construction of multi-dimensional feature database:

[0036] Label the images according to 9 types of thermal injury, 86 anatomical divisions, and 4 stages of injury degree;

[0037] Store the feature association relationship by using a graph database; construct a human burn trace database module;

[0038] S3. Multi-modal feature extraction:

[0039] Extract the texture fractal dimension of the skin charring area, the edge gradient distribution of the blister shape, the color space difference between the wound and the normal tissue, and the infrared thermal radiation features of the deep tissue injury by using a convolutional neural network (CNN);

[0040] S4. Cascade deep learning model training:

[0041] Divide the training set, validation set, and test set according to the ratio of 70:15:15;

[0042] Train and execute in sequence:

[0043] Heat damage trace type recognition and positioning, i.e. damage type classification: an EfficientNetB4 model is used to output a 9-class probability distribution;

[0044] Injury process reconstruction: an LSTM network is used to generate a time-intensity curve;

[0045] Death state discrimination: a Transformer model is used to fuse three types of biomarkers;

[0046] Death mode determination: a probability matrix of suicide, homicide or accident is generated;

[0047] S5. Intelligent research and report generation:

[0048] Input new case images and match feature templates through cosine similarity > 0.85;

[0049] Generate a comprehensive report, including:

[0050] Injury mechanism animation;

[0051] 95% confidence interval of death time;

[0052] 0-100 score of arson suspect index;

[0053] Generate a key evidence attribution heat map;

[0054] S6. Explainable output:

[0055] Generate a decision path tree diagram, with the path being: carbonization fractal dimension > 0.7 → water bubble gradient variance < 2.5 → CO-Hb > 60% → determine antemortem burning.

[0056] Further, in step S4, the cascaded deep learning model needs to meet the following conditions:

[0057] When the probability distribution output by the heat damage trace type recognition and positioning has a confidence level of the "mechanical superimposed damage" category higher than a preset threshold, activate the spatiotemporal correlation module of skeletal fracture and surface burn for analysis;

[0058] When the probability distribution output by the heat damage trace type recognition and positioning belongs to any one of the categories of combustion-supporting agent contact type heat damage, combustion-supporting agent deflagration radiation type heat damage, or combustion-supporting agent heat damage post-healing trace, with a confidence level higher than a preset threshold, start the cross-modal attention mechanism to correlate the image and chemical spectrum for analysis.

[0059] Furthermore, the spectral data and microscopic image data of the cardiac CO-Hb concentration spectral analysis and respiratory carbon deposition analysis combined in the death status determination in step S4 are derived from forensic toxicology and pathology examination reports.

[0060] Furthermore, in step S5, the generated key evidence attribution heatmap is generated using the Grad-CAM algorithm; in step S6, a decision path tree diagram of key decision factors is generated based on the SHAP value.

[0061] By adopting the above technical solution, the present invention has at least one of the following beneficial effects compared with the prior art:

[0062] This invention solves the accuracy and efficiency problems caused by the reliance on subjective experience in traditional fire investigations, providing scientific and efficient technical support for case solving. It reconstructs the causal chain from microscopic traces to the macroscopic nature of the fire (accident / malicious / arson), providing quantitative technical evidence for forensic identification. Attached Figure Description

[0063] Figure 1 This is a technical flowchart of the present invention, illustrating the entire process of data acquisition, database construction, trace feature extraction, deep learning model training, intelligent trace comparison and application. Detailed Implementation

[0064] The following is in conjunction with the appendix Figure 1 The present invention will be further described in detail below to facilitate a clear understanding of the invention, but these descriptions do not constitute a limitation thereof.

[0065] Example 1

[0066] like Figure 1 As shown in this embodiment, an intelligent analysis system based on thermal injury traces of the human body in a fire is a system for intelligently analyzing the causes of death and injury and the nature of the fire based on thermal injury traces of the human body in a fire. Specifically, it includes the following modules in sequence:

[0067] Image acquisition module: used to acquire digital images of burn marks on the bodies of the dead and injured in the fire;

[0068] Human burn scar database module: used to store annotated burn image datasets; the annotation dimensions of the human burn scar database module include:

[0069] The heat damage type label: flame heat damage, hot body contact heat damage, gas explosion heat damage, combustion-supporting agent contact heat damage, combustion-supporting agent explosion radiation heat damage, combustion-supporting agent heat damage post-healing trace, mechanical superimposed damage, and pre-death state burn; wherein the combustion-supporting agent contact heat damage and the combustion-supporting agent explosion radiation heat damage both include combustion-supporting agent residual form identification and explosion radiation range modeling, which provide strong support for fire cause analysis.

[0070] The anatomical positioning label of the damage site: 86 key regions according to the international anatomical terminology; the damage time and degree associated label: divided into I hyperemia period to IV carbonization period according to histological changes.

[0071] The feature extraction module: a convolutional neural network (CNN) is used to extract multi-modal morphological features from the input image; the feature extraction module includes:

[0072] The texture fractal dimension of the skin carbonization area;

[0073] The edge gradient distribution of the blister shape;

[0074] The color space difference between the wound and normal tissue;

[0075] The infrared thermal radiation characteristics of deep tissue damage.

[0076] The intelligent research and judgment module: based on a cascaded deep learning model architecture, sequentially performs:

[0077] (a) heat damage trace type identification and positioning: positioning the specific heat damage type area through a spatial attention mechanism;

[0078] (b) injury process reconstruction: relying on a time series recurrent network to simulate the damage formation dynamics process;

[0079] (c) death state discrimination: combining respiratory tract carbon deposition, heart blood CO-Hb concentration spectral analysis, and skin tissue elasticity characteristics to output the probability value of pre-death burning or post-death cremation; the death state discrimination in the intelligent research and judgment module is achieved by analyzing respiratory tract carbon deposition, heart blood carbon monoxide hemoglobin concentration, and skin tissue elasticity.

[0080] (d) death mode determination: generating a probability matrix of suicide, homicide, or accident;

[0081] (e) comprehensive determination of fire nature: integrating the output results of (a)-(d) to generate a probability assessment of arson, accident, or natural disaster.

[0082] In this embodiment, the hardware deployment of the intelligent analysis system also includes:

[0083] The edge computing node: used to realize on-site image preprocessing;

[0084] Central server: for running multi-task inference model;

[0085] Forensic interaction terminal: for displaying timeline deduction animation.

[0086] Embodiment 2

[0087] This embodiment is an intelligent analysis method based on human thermal injury traces in fire scenes, which adopts the intelligent analysis system of human thermal injury traces in fire scenes in Embodiment 1, and includes the following steps:

[0088] S1: data acquisition and standardization processing:

[0089] Collect no less than 100 cases of burn images identified by forensic experts, and label the injury type, anatomical position, and injury mechanism of each image;

[0090] Perform thin plate spline transformation to correct the distortion of the corpse posture, use I-Net network to segment the adhesion injury area, and form an anatomical positioning reference map;

[0091] S2: multi-dimensional feature database construction:

[0092] According to 9 types of thermal injury, 86 anatomical divisions, and 4 levels of injury severity staging, label the images;

[0093] Use a graph database to store the feature association relationship; and construct a human burn trace database module;

[0094] S3. Multi-modal feature extraction:

[0095] Extract the texture fractal dimension of the skin charring area, the edge gradient distribution of the blister shape, the color space difference between the wound and the normal tissue, and the infrared thermal radiation features of the deep tissue injury through the convolutional neural network CNN;

[0096] S4. Cascade deep learning model training:

[0097] Divide the training set, validation set, and test set in the ratio of 70:15:15;

[0098] Train and execute in sequence:

[0099] Thermal injury trace type recognition and positioning, i.e. injury type classification: use the EfficientNetB4 model to output 9 types of probability distribution;

[0100] Injury process reconstruction: use the LSTM network to generate a time-intensity curve;

[0101] Death state determination: a Transformer model is used to integrate three types of biomarkers; the CO-Hb concentration spectrum analysis and respiratory tract carbon deposition analysis in the death state determination in step S4 are combined, and the corresponding spectral data and microscopic image data come from the forensic toxicology and pathology test report.

[0102] Death mode determination: generate a probability matrix for suicide, homicide, or accident;

[0103] The cascade deep learning model needs to meet the following conditions:

[0104] When the confidence of the "mechanical superimposed damage" category in the probability distribution of the thermal damage trace type recognition and positioning output is higher than the preset threshold, the spatiotemporal correlation module of bone fracture and surface burn is activated for analysis;

[0105] When any of the confidence of the combustion-supporting agent contact type thermal damage, combustion-supporting agent explosion radiation type thermal damage, or combustion-supporting agent thermal damage post-healing trace category in the probability distribution of the thermal damage trace type recognition and positioning output is higher than the preset threshold, the cross-modal attention mechanism is started to correlate the image and chemical spectrum for analysis.

[0106] S5. Intelligent research and report generation:

[0107] Input new case images and match feature templates through cosine similarity with a value greater than 0.85;

[0108] Generate a comprehensive report, including:

[0109] Injury mechanism deduction animation;

[0110] 95% confidence interval of death time;

[0111] 0-100 point arson suspect index;

[0112] Generate a key evidence attribution heat map through Grad-CAM algorithm;

[0113] S6. Explainable output:

[0114] Generate a decision path tree graph based on SHAP values, with the path being: carbonization fractal dimension > 0.7 → water bubble gradient variance < 2.5 → CO-Hb > 60% → determine antemortem burning.

[0115] Example 3 (arson case analysis)

[0116] Case background: a fire broke out in a residential building, causing the death of two people. The police initially suspected it was an arson case, and the corpses of the deceased needed to be analyzed to obtain evidence.

[0117] Data collection and standardization: Collect 15 high-definition photos of burn cases of 2 corpses at the fire scene from different angles and parts, label the damage type (such as deep burn, superficial burn, etc.), anatomical location (such as face, limbs, trunk, etc.), and injury mechanism (initially speculated as flame burn) of each image. Correct the posture distortion of the corpse caused by the environment of the fire scene (such as the corpse curling up, etc.) using thin plate spline transformation, segment the adhesion damage area (such as the part of the skin and clothes adhered after burn) using U-Net network, form an anatomical positioning reference map, and clearly show the relative position relationship between the damage of each part of the corpse and the normal tissue.

[0118] Multi-dimensional feature database construction: According to the 9 types of thermal damage, 86 anatomical divisions, and 4 levels of damage degree staging standards in Example 2, the corpse image of the case is labeled in detail. For example, in the facial anatomical division (which belongs to one of the 86 anatomical divisions), superficial burn (which belongs to one of the 9 types of thermal damage) is found, and the damage degree is 1 (mild). Store the correlation of these labeled data using graph database, and construct the human burn trace database module to facilitate subsequent model training and feature extraction.

[0119] Multi-modal feature extraction: Use convolutional neural network (CNN) to extract the texture fractal dimension of the carbonized skin area (such as the fractal dimension of the facial skin carbonized area is 0.85), the edge gradient distribution of the blister shape (such as the edge gradient variance of the limb blister is 2.0), the color space difference between the wound and the normal tissue (such as the difference value between the wound and the normal tissue in the RGB color space of the trunk is 45), and the infrared thermal radiation characteristics of the deep tissue damage (such as the infrared thermal radiation intensity peak value of the muscle tissue is 320).

[0120] Cascade deep learning model training: Divide the large amount of data collected earlier, including arson cases and non-arson cases, into training set, validation set, and test set in the ratio of 70:15:15. First, use the training set to train the thermal damage trace type recognition and positioning model, use the EfficientNetB4 model to recognize the corpse image of the case, and output 9 probability distributions, where the probability of the "flame contact type thermal damage" category is 0.8 (higher than the preset threshold 0.7). At this time, activate the spatio-temporal association module of bone fracture and surface burn to analyze the correlation of bone damage and surface burn in time and space, and further search for arson evidence.

[0121] Intelligent analysis and report generation: Input the new case image, the system matches to the arson case feature template in the previous database through the cosine similarity of >0.85 (similarity of 0.89). The generated comprehensive report contains injury mechanism deduction animation, showing the process of fire spreading from the arson point to causing burns on different parts of the body; the 95% confidence interval for the time of death is 10-15 minutes after the fire; the arson suspect index is 92 points (0-100 points). Generate a key evidence attribution heat map through the Grad-CAM algorithm, showing that the thermal damage trace features of the face and hands have a key impact on determining the arson case, and these parts of the burn trace are highly related to the splashing of accelerants and direct contact with fire during arson.

[0122] Interpretable output: Based on SHAP values, generate a decision path tree diagram, the path is: charred fractal dimension >0.7 (face charred fractal dimension is 0.85) → blister gradient variance <2.5 (limb blister gradient variance is 2.0) → CO-Hb >60% (CO-Hb concentration detected in heart blood is 68%) → determine that the person was alive at the time of burning, and combined with evidence such as on-site accelerant residue detection, further support the determination of the arson case.

[0123] Example 4 (Accident identification)

[0124] Case background: A fire occurred in a factory workshop, resulting in injury to one worker. It needs to be determined whether the fire was caused by a production accident and whether the worker's injuries are consistent with the accident.

[0125] Data collection and standardization: Collect burn case images of the injured worker (a total of 10 photos from different angles and parts), label the injury type (such as partial-thickness burn), anatomical location (such as arm, leg, etc.), and injury mechanism (initially suspected to be a flame burn caused by high-temperature equipment). Perform thin plate spline transformation to correct the posture distortion of the corpse due to the irregular posture in the factory workshop environment (such as the twisted posture when the worker avoids the fire), use the U-Net network to segment the adhesion damage area (such as the part of the skin adhered to the factory protective clothing after burning), and form an anatomical positioning reference map to clearly show the relative position of the worker's injuries and normal tissues in each part.

[0126] Multi-dimensional feature database construction: Label the worker's corpse image according to the classification standards in Example 2, for example, in the arm anatomical partition (belonging to one of the 86 anatomical partitions), partial-thickness burn (belonging to one of the 9 types of thermal damage) is found, and the injury degree is 2 (moderate). Store the feature association relationship through the graph database to construct the human burn trace database module, which is convenient for subsequent model calling and analysis.

[0127] Multi-modal feature extraction: The texture fractal dimension of the charred area of the skin (e.g., the fractal dimension of the charred area of the arm skin is 0.78), the edge gradient distribution of the blister shape (e.g., the leg blister edge gradient variance is 3.0), the color space difference between the wound and normal tissue (e.g., the difference value between the arm wound and normal tissue in the RGB color space is 38), and the infrared thermal radiation characteristics of the deep tissue damage (e.g., the infrared thermal radiation intensity peak of the subcutaneous fat tissue is 280) are extracted using the convolutional neural network (CNN).

[0128] Cascade deep learning model training: Similarly, the previously collected data containing various fire cases is divided into training set, validation set and test set according to the ratio of 70:15:15. The training set is used to train the heat damage trace type identification and positioning model. The EfficientNetB4 model is used to identify the worker image, and the output is a probability distribution of 9 categories. The confidence of the "mechanical superimposed damage" category is 0.6 (lower than the preset threshold of 0.7), and the related analysis module is not activated. The confidence of the "flame burn" category is 0.85 (higher than the preset threshold of 0.7), and the subsequent analysis is continued. The injury process reconstruction model uses the LSTM network to generate a time-intensity curve, showing the change of fire intensity from the fire to the worker injury process. The death state discrimination model uses the Transformer model to fuse three types of biomarkers (including CO-Hb concentration, blood inflammation markers, and tissue damage markers), combines spectral data and microscopic image data, and judges that the worker is currently in an injured but not dead state. The death mode determination model generates a probability matrix of suicide, homicide or accident, and the result shows that the accident probability is 98%.

[0129] Intelligent research and report generation: Input the new case image, the system matches the production accident fire feature template in the database through the cosine similarity of >0.85 (the similarity is 0.88). The generated comprehensive report contains an injury mechanism deduction animation, which shows the process of the worker being burned by the flame during the operation of the equipment when the fire occurs in the factory workshop; the injury time interval with a confidence of 95% is 5-8 minutes after the fire; the arson suspect index is 8 points (0-100 points), which basically excludes the arson suspect. The key evidence attribution heat map is generated by the Grad-CAM algorithm, which shows that the burn marks on the worker's arms and legs are the key evidence for determining the production accident fire, and these parts of the burn are highly consistent with the height and spread path of the flame when the equipment in the factory workshop caught fire.

[0130] The explainable output: based on the SHAP value, a decision path tree diagram is generated, the path is: carbonization fractal dimension > 0.7 (the arm carbonization fractal dimension is 0.78) -> blister gradient variance < 2.5 (the leg blister gradient variance is 3.0, which is slightly higher than the threshold value, but the judgment is still an accident in combination with other characteristics) -> CO-Hb is 45% (the upper limit of the normal range is generally 5%, but the worker is not dead, and is in the fire scene, there is a certain inhalation of carbon monoxide) -> determine as an injury caused by a production accident fire, which provides a strong basis for subsequent accident liability determination and work injury compensation, etc.

[0131] The above is only a preferred embodiment of the present application, and does not limit the structure of the present application in any form. The arrangement type and the number of uses of the present application are not limited to the example, and can be optimized according to the actual engineering. Any modification, equivalent change and decoration of the above embodiment according to the technical principle of the present application, which does not deviate from the technical scheme of the present application, is still within the scope of the technical scheme of the present application.

Claims

1. An intelligent analysis system based on thermal injury traces of human bodies in fires is a system for intelligently analyzing the causes of death and injury and the nature of fires based on thermal injury traces of human bodies in fires. Its characteristics are: Specifically, it includes the following modules in sequence: Image acquisition module: used to acquire digital images of burn marks on the bodies of the dead and injured in the fire; Human burn scar database module: used to store labeled burn image datasets; Feature extraction module: Employs a convolutional neural network (CNN) to extract multimodal morphological features from the input image; Intelligent judgment module: Based on a cascaded deep learning model architecture, it executes sequentially as follows: (a) Identification and localization of thermal damage trace types: Localizing regions with specific thermal damage types through spatial attention mechanisms; (b) Injury process reconstruction: The dynamic process of injury formation is simulated based on temporal recursive networks; (c) Determination of death status: Combining respiratory carbon deposition, cardiac blood CO-Hb concentration spectrum analysis and skin tissue elasticity characteristics, output the probability value of burning to death before life / cremation after death; (d) Determining the manner of death: Generate a probability matrix for suicide, homicide, or accident; (e) Comprehensive determination of fire nature: Combine the output results of (a)-(d) to generate a probability assessment of arson, accident, and natural disaster.

2. The intelligent analysis system based on thermal injury traces of the human body in a fire, as described in claim 1, is characterized in that... The annotation dimensions of the human burn scar database module include: Thermal damage type tags: flame thermal damage, incandescent body contact thermal damage, gas explosion thermal damage, oxidizer contact thermal damage, oxidizer deflagration radiation thermal damage, oxidizer thermal damage healing marks, mechanical superposition damage, near-death burns; Anatomical location labels for the injury site: 86 key areas defined by international anatomical terminology; Injury time and severity correlation label: divided into stage I (congestion stage) to stage IV (carbonization stage) according to histological changes.

3. The intelligent analysis system based on thermal injury traces of the human body in a fire, as described in claim 2, is characterized in that... The feature extraction module includes: The fractal dimension of texture in carbonized areas of the skin; Edge gradient distribution of bubble morphology; Color space differences between wound and normal tissue; Infrared thermal radiation characteristics of deep tissue damage.

4. The intelligent analysis system based on thermal damage traces of the human body in a fire, as described in claim 3, is characterized in that... Both the contact-type thermal damage and the deflagration-radiation-type thermal damage of the oxidizer include: identification of the residual morphology of the oxidizer and modeling of the deflagration radiation range, providing strong support for fire cause analysis.

5. The intelligent analysis system based on thermal injury traces of the human body in a fire, as described in claim 4, is characterized in that, The death status determination in the intelligent judgment module is achieved by analyzing carbon deposits in the respiratory tract, carboxyhemoglobin concentration in the cardiac blood, and skin tissue elasticity.

6. The intelligent analysis system based on thermal injury traces of the human body in a fire, as described in claim 1, is characterized in that, The hardware deployment of the intelligent analysis system also includes: Edge computing nodes: used to perform on-site image preprocessing; Central server: Used to run multi-task inference models; Forensic interactive terminal: used to display timeline simulation animations.

7. An intelligent analysis method based on thermal injury traces of the human body in a fire, comprising the intelligent analysis system based on thermal injury traces of the human body in a fire as described in any one of claims 1-6, characterized in that, Includes the following steps: S1: Data Acquisition and Standardization Processing Collect images of no fewer than 100 forensic burn cases and label the injury type, anatomical location, and injury mechanism for each image; Thin-plate spline transformation was performed to correct cadaver posture distortion, and a U-Net network was used to segment adhesion and injury areas to create an anatomical localization reference. S2: Construction of a multi-dimensional feature database: Images are labeled according to 9 types of thermal injury, 86 anatomical regions, and 4 levels of injury severity. A graph database is used to store feature relationships; a human burn scar database module is constructed. S3. Multimodal Feature Extraction: The texture fractal dimension of the skin carbonization area, the edge gradient distribution of blister morphology, the color space difference between the wound and normal tissue, and the infrared thermal radiation characteristics of deep tissue damage were extracted by using a convolutional neural network (CNN). S4. Training of Cascaded Deep Learning Models: The training set, validation set, and test set are divided into three groups according to a 70%:15%:15% ratio. Train sequentially and execute in cascades: Thermal damage trace type identification and localization, i.e. damage type classification: The EfficientNetB4 model is used to output 9 probability distributions; Injury process reconstruction: Time-intensity curves were generated using an LSTM network; Mortality status determination: A Transformer model is used to fuse three types of biomarkers; Method of death determination: Generate a probability matrix for suicide, homicide, or accident; S5. Intelligent Analysis and Report Generation: Input a new case image and match feature templates using a cosine similarity greater than 0.85; Generate a comprehensive report, including: Animation demonstrating the injury mechanism; 95% confidence level time interval for death; Arson suspicion index: 0-100 points; Generate a heatmap of key evidence attribution; S6. Interpretable output: Generate a decision path tree diagram. The path is: carbonization fractal dimension > 0.7 → water bubble gradient variance < 2.5 → CO-Hb > 60% → determine ante-death by burning.

8. The intelligent analysis method based on thermal damage traces of the human body in a fire, as described in claim 6, is characterized in that... In step S4, the cascaded deep learning model must meet the following conditions: When the confidence level of the "mechanical superposition damage" category in the probability distribution of the thermal damage trace type identification and localization output is higher than the preset threshold, the spatiotemporal correlation module of bone fracture and surface burn is activated for analysis. When the probability distribution of the thermal damage trace type identification and localization output shows that any of the categories of oxidizer contact thermal damage, oxidizer deflagration radiation thermal damage, or oxidizer thermal damage healing traces has a confidence level higher than a preset threshold, a cross-modal attention mechanism is activated to associate the image with the chemical spectrum for analysis.

9. The intelligent analysis method based on thermal damage traces of the human body in a fire, as described in claim 8, is characterized in that... The cardiovascular CO-Hb concentration spectral analysis and respiratory carbon deposition analysis combined in the death status determination in step S4 are based on spectral data and microscopic image data obtained from forensic toxicology and pathology examination reports.

10. The intelligent analysis method based on thermal damage traces of the human body in a fire, as described in claim 9, is characterized in that... In step S5, the generated key evidence attribution heatmap is generated using the Grad-CAM algorithm; in step S6, a decision path tree diagram of key decision factors is generated based on the SHAP value.

Images