A survey system for forest fire disaster post-investigation

Abstract

This invention relates to the field of forest fire investigation technology and discloses an exploration system for post-fire investigation and evidence collection, including a handheld acquisition terminal, a drone mapping terminal, a data aggregation node, and a backend system. The handheld acquisition terminal collects trace images and has a manual correction module for on-site personnel to correct identification results. The data aggregation node is used to build a dedicated 2.4GHz wireless network in weak network environments and supports multi-hop relay. The backend system includes a task coordination and push module, which generates mapping instructions based on the marked images uploaded by the handheld terminal, driving the drone to prioritize close-up detailed investigation of the marked areas. The manually corrected data is fed back to the model training module for iterative optimization. This invention achieves a real-time collaborative closed loop of human marking and drone re-examination in post-disaster investigation scenarios, continuously improving identification accuracy and on-site investigation efficiency, and effectively solving the problem of real-time data interaction in public network blind spots.

Description

Technical Field

[0001] This invention relates to the field of forest fire investigation technology, specifically to an exploration system for post-fire investigation and evidence collection in forest fires. Background Technology

[0002] Forest fires are among the most significant natural disasters globally, and forest fire investigations are crucial for analyzing fire causes, determining liability, and developing preventative measures. After a fire breaks out, investigators must thoroughly examine the scene to collect evidence, analyze the direction of fire spread, and trace the origin of the fire. However, forest fire sites are typically located in remote mountainous areas with complex terrain, making on-site investigations difficult, inefficient, and challenging to identify typical traces. Public network communication signals are often insufficient, making real-time data exchange and collaborative investigations extremely difficult.

[0003] Existing forest fire-related technologies have the following main shortcomings:

[0004] 1) Lack of a system-level solution for post-disaster investigation and evidence collection: Existing technologies are mostly focused on early fire warning or firefighting decision-making during a disaster, with the core objective of "detecting the fire" or "assisting in firefighting." However, for post-disaster investigation, manual surveys are time-consuming and difficult to quickly cover large areas of the fire scene; at the same time, the terrain at the fire scene is complex and there may be dangers such as residual fires and landslides, which pose a threat to the safety of the survey personnel.

[0005] 2) Existing emergency communication solutions are incompatible with investigation and evidence collection scenarios: In the field of mine emergency communication, a multi-hop relay transmission method based on the integration of LoRa and 4G (such as CN121531336A) has been proposed for uploading personnel vital signs or environmental parameters in communication blind spots. However, this solution is geared towards nodes automatically triggering emergency data uploads, with the goal of "transmitting the data," rather than the business process of investigators actively marking suspected traces and directing drones for re-examination and evidence collection.

[0006] 3) Exploration and mapping are independent, resulting in low collaboration efficiency: When multiple personnel conduct collaborative exploration, information transmission is not timely, making efficient collaboration difficult. Ground exploration and UAV mapping are usually carried out separately. After exploration personnel discover suspected traces, it is difficult to promptly direct UAVs for close-up and detailed investigation, thus missing the best opportunity to collect evidence.

[0007] 4) Insufficient accuracy: On-site identification results and back-end identification results cannot be combined and corrected. Back-end intelligent identification results are mostly pushed one-way. Although some studies involve methods for automatically adjusting model parameters based on observation data, the adjustment logic is an automatic iteration within the system, without incorporating the professional judgment of on-site investigators. Furthermore, relying solely on manual judgment of the fire spread direction and ignition source location is easily affected by subjective factors, making it difficult to guarantee accuracy.

[0008] 5) Lack of professional identification capabilities for typical post-disaster traces: Current technologies identify targets that are burning or about to burn, such as smoke, fire lines, temperature, and smoke. There is no systematic intelligent solution for professional identification of post-disaster traces—such as determining the direction of spread based on vegetation burn morphology and the tilt angle of the charred layer, identifying the electric ignition source based on the characteristics of electrical melting marks, identifying the lightning ignition source based on the characteristics of lightning strike melting marks, and identifying traces of electrical burns and the spread of burns based on burn characteristics.

[0009] In summary, while some existing fire investigation systems are based on drones and image recognition, their functions are relatively limited, lacking intelligent recognition of fire traces and support for multi-person collaboration. Furthermore, they struggle to achieve real-time data transmission and processing in environments with poor network conditions. There is an urgent need for an intelligent investigation system designed for post-disaster investigation and evidence collection scenarios. This system should enable real-time collaboration between ground surveys and aerial mapping, support on-site personnel to actively mark and drive drone resurveys, possess manual correction and model iterative learning capabilities, and be able to professionally identify post-disaster traces. Summary of the Invention

[0010] To address the aforementioned problems, the present invention aims to provide a survey system for post-fire investigation and evidence collection in forest fires. This system can collect typical combustion spread traces and ignition source traces at the fire scene using handheld devices and drones, and utilize a backend system for intelligent identification and processing. This enables rapid determination of the fire spread direction, precise location of the ignition source, and efficient support for multi-person collaborative surveys. The technical solution is as follows:

[0011] An exploration system for post-fire investigation and evidence collection in forest fires, comprising:

[0012] Handheld acquisition terminal: used to collect trace images containing geographic coordinates and sensor information on site and send them to the back-end system, while receiving and displaying the analysis results of the back-end system; the handheld acquisition terminal supports switching between intranet direct connection mode and extranet cloud server forwarding mode, and automatically switches to extranet cloud server for data transmission when intranet connection fails;

[0013] Unmanned aerial vehicle (UAV) mapping terminal: used to conduct aerial mapping of fire scenes to obtain multispectral image data for constructing 3D models of the scene;

[0014] Data aggregation nodes include 4G / 5G network routers and wireless power amplifiers, used to build a dedicated 2.4GHz wireless network that supports real-time local interaction among multiple terminals in offline or weak network environments, receive and cache data uploaded by the handheld acquisition terminals, and interact with the backend system; long-distance data transmission is relayed through a multi-hop network composed of multiple data aggregation nodes.

[0015] The backend system includes:

[0016] Trace recognition module: used to analyze and process the received trace images, identify the direction of fire spread, the type of ignition source and the type of burn marks in the images, and output the recognition results;

[0017] Spatial analysis and plotting module: used to receive multispectral image data from the UAV mapping terminal and overlay and fuse it with the electronic map of the geographic information system; combined with the recognition results output by the trace recognition module and its corresponding geographic coordinates, dynamically plot the vector arrow of the fire spread trend and the hot spot area of ​​the ignition source on the electronic map;

[0018] Task collaboration and push module: running in the background system, when a new image containing suspected spread traces or ignition source traces is received from a handheld acquisition terminal, the module analyzes the geographical coordinate range of the marker, generates a high-priority mapping instruction and sends it to the UAV mapping terminal, so that it can prioritize close-up detailed investigation of the marked area.

[0019] Manual correction module: running on the handheld acquisition terminal, used to receive and overlay the recognition results pushed back by the backend system on the display screen, and allow on-site personnel to manually confirm or correct the recognition results based on their professional knowledge, and return the correction information to the backend system.

[0020] The beneficial effects of this invention are:

[0021] 1. Specialized design for post-disaster investigation and evidence collection: This invention is specifically designed for fire investigation scenarios. The objects to be identified include post-disaster traces such as the tilt angle of the carbonized layer, electric melting marks, and lightning strike melting marks, filling the gap in existing technologies that focus on pre-disaster early warning and in-disaster firefighting.

[0022] 2. Real-time closed-loop human-machine collaboration: This invention realizes a real-time task closed loop of "on-site marking by surveyors → background analysis → priority re-survey by UAVs" through a task collaboration and push module, shortening the time difference between ground discovery and aerial re-survey to the minute level, and realizing real-time human-machine collaboration in post-disaster investigation scenarios for the first time.

[0023] 3. Model Iterative Calibration: This invention integrates the professional knowledge of field investigators into the model optimization process through the linkage of the manual correction module and the model training module, realizing a closed-loop iteration of "machine recognition → human confirmation → machine learning", resulting in high recognition accuracy;

[0024] 4. Communication assurance in weak / no network environments: This invention constructs a dedicated 2.4GHz wireless network through data aggregation nodes, supports multi-hop relay, and, in conjunction with the automatic switching mechanism between the internal and external networks of handheld terminals, ensures real-time data transmission and multi-terminal collaboration even in public network blind spots such as deep mountains and dense forests.

[0025] 5. Multi-source data fusion and visualization: This invention integrates multi-source data such as UAV 3D modeling data, GIS electronic maps, trace recognition results, and manual correction records, and dynamically plots the spread trend arrows and ignition source hotspot areas on the electronic map. It supports clicking on icons to view the original photos, providing investigators with intuitive decision support. Attached Figure Description

[0026] Figure 1 This is a basic framework diagram of the exploration system for post-forest fire investigation and evidence collection according to the present invention.

[0027] Figure 2 This is a flowchart illustrating the workflow of the exploration system for post-forest fire investigation and evidence collection according to the present invention. Detailed Implementation

[0028] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0029] I. System Composition and Hardware Configuration:

[0030] This embodiment provides a survey system for post-forest fire investigation and evidence collection, including a handheld data acquisition terminal, a drone mapping terminal, a data aggregation node, and a backend system. The basic framework diagram is shown below. Figure 1 As shown.

[0031] 1. Handheld data acquisition terminal:

[0032] The handheld acquisition terminal is used to collect trace images containing geographic coordinates and sensor information on site, send them to the backend for processing, and receive and display the analysis results from the backend. The handheld acquisition terminal supports switching between intranet direct connection mode and extranet cloud server forwarding mode, and automatically switches to extranet cloud server for data transmission when the intranet connection fails. Voice intercom communication is supported between multiple handheld terminals and between the handheld terminals and the backend system.

[0033] The handheld data acquisition terminal features an industrial-grade rugged design, making it suitable for complex outdoor environments. Its hardware configuration includes: a high-performance processor, a high-resolution camera (≥1080P), a GPS / BeiDou dual-mode positioning module, a gravity sensor, a direction sensor, a 4G / 5G communication module, a 2.4GHz wireless communication module, a voice intercom module, and a touchscreen display.

[0034] Handheld data acquisition terminals include:

[0035] (1) Multi-sensor fusion imaging module, used to simultaneously trigger gravity sensor, orientation sensor and global positioning system when collecting trace images, and encapsulate the image at the moment of shooting, precise geographical coordinates, shooting angle and device attitude information into structured data packets; supports automatic focusing when taking pictures, and saves the image to local storage first after taking pictures.

[0036] When the multi-sensor fusion imaging module of the handheld data acquisition terminal is in operation, surveyors trigger the image capture command via the touchscreen. Simultaneously, the camera captures trace images while the gravity sensor, orientation sensor, and GPS are activated. The system encapsulates the image at the moment of capture, geographical coordinates (longitude and latitude), shooting angle (equipment orientation azimuth), and equipment attitude information (roll and pitch angles) into a structured data packet. After capturing the image, it is prioritized for local storage to ensure no data loss.

[0037] (2) Communication module, responsible for data uploading and receiving, supports uploading the structured data packets through 4G / 5G public network or 2.4GHz dedicated wireless link established with the data aggregation node, and receives analysis results from the background system; the communication module automatically displays the intranet or extranet connection status according to the network connection status, and provides photo retransmission function in case of network outage in the historical photo interface.

[0038] The handheld data acquisition terminal supports automatic switching between intranet direct connection mode and external cloud server forwarding mode. When the device detects that the 2.4GHz dedicated wireless network connection established with the data aggregation node is normal, it prioritizes the intranet direct connection mode for data transmission; when the 2.4GHz network is unavailable, but the 4G / 5G public network signal is good, it automatically switches to the external cloud server forwarding mode; when both networks are unavailable, the data is temporarily stored locally and automatically retransmitted after the network is restored.

[0039] When within the coverage area of ​​a dedicated 2.4GHz wireless network, it communicates with the data aggregation node through this network; when in a network dead zone, it uploads data via the 4G / 5G public network. The communication module provides a photo retransmission function in case of network outage in the historical photo interface. Surveyors can view the list of uploaded and unuploaded photos and manually trigger retransmission.

[0040] (3) Result visualization and correction module, used to receive and overlay the fire spread direction arrow and the predicted range of the ignition source pushed back by the background system on the display screen, and to allow on-site personnel to manually confirm or correct the identification results. The correction information is returned to the background system through the communication module as sample data of the model training module.

[0041] After receiving the identification results pushed by the backend system, the manual correction module of the handheld acquisition terminal overlays an arrow indicating the direction of fire spread and the predicted range of the ignition source on the display screen. Surveyors can manually confirm or correct the identification results. For example, if the spread direction identified by the system does not match actual observation, surveyors can rotate the direction arrow on the touchscreen to make corrections; if the location of the ignition source marked by the system is inaccurate, the marker can be dragged to correct its position. The correction information is returned to the backend system via the communication module.

[0042] (4) Voice intercom module, including one-click call function, used to receive or initiate group voice communication with the background system and other handheld acquisition terminals through the communication module; after the survey personnel press the intercom button, they can initiate group voice communication to the background system or other online handheld acquisition terminals to realize real-time communication in multi-soldier collaborative operations.

[0043] (5) System configuration module, used to configure device number, server number, external server IP address and port, internal server IP address and port, and supports camera parameter settings, including resolution, white balance and exposure parameters.

[0044] 2. Unmanned Aerial Vehicle (UAV) mapping terminal:

[0045] The drone mapping terminal is used to conduct aerial mapping of fire scenes to obtain multispectral image data for building three-dimensional models of the scene, and supports real-time transmission of video streams to the back-end system for real-time capture via real-time streaming protocols.

[0046] UAV mapping terminals include:

[0047] (1) Autonomous task planning module, which is used to automatically generate a fine planar route mapping task covering the target area after receiving the mapping instruction issued by the task coordination and push module; supports setting mapping accuracy, flight altitude mode and digital surface model file; has power monitoring and automatic return function, and can automatically resume the unfinished mapping task after replacing the battery.

[0048] (2) Multispectral remote sensing module, which integrates a visible light high-definition camera and a thermal infrared imager, is used to simultaneously collect visible light images and surface thermal radiation distribution data of the fire scene, provide auxiliary data for the identification of hidden fire sources, and provide basic image data for the background three-dimensional modeling;

[0049] (3) Real-time image transmission module, which supports the real-time transmission of video streams collected by the drone camera to the back-end system via RTSP (Real-time Streaming Protocol) for back-end operators to observe and capture images in real time.

[0050] The UAV mapping terminal utilizes a multi-rotor industrial-grade UAV platform, equipped with a multispectral remote sensing module, and integrates a visible light high-definition camera (≥4K) and a thermal infrared imager. The UAV features an autonomous mission planning module that, upon receiving mapping instructions from the backend system, automatically generates a detailed planar flight path mapping task covering the target area, supporting settings for mapping accuracy, flight altitude mode, and digital surface model files. The UAV also features battery monitoring and automatic return-to-home functionality; when the battery level falls below a preset threshold, it automatically returns to the takeoff point. After battery replacement, it can automatically resume unfinished mapping tasks without needing to replan the flight path.

[0051] 3. Data aggregation node:

[0052] The data aggregation node is used to build a dedicated 2.4GHz wireless network in offline or weak network environments, receive and cache the data uploaded by the handheld acquisition terminal, and interact with the backend system; long-distance data transmission is achieved by forming a multi-hop network through multiple data aggregation nodes for data relay.

[0053] The data aggregation node consists of a 4G / 5G network router and a wireless power amplifier (AP power amplifier). The 4G / 5G network router, equipped with a SIM card, connects to the public network, providing internet access for the aggregation node. The AP power amplifier, connected to the router via a wired connection, amplifies the 2.4GHz wireless signal, extending wireless coverage (up to 2km or more in open conditions) and supports multi-level bridging to further extend communication distance. The AP power amplifier is configured in relay bridging mode, capable of scanning and connecting to upstream wireless networks while providing a local 2.4GHz wireless access point. Multiple data aggregation nodes relay data through a multi-hop network, ensuring long-distance data transmission in complex terrain.

[0054] 4. Backend System:

[0055] The backend system is deployed on a server and includes modules for project and equipment management, data interface, model training, trace recognition, spatial analysis and plotting, task collaboration and push, case management and filing, and time-series synchronization engine.

[0056] (1) The project and equipment management module is used to manage multiple fire investigation projects, support the creation of new projects, setting of current projects, and export of project backups; at the same time, it registers and manages handheld acquisition terminal devices, maintains the online device list and its status, and provides device routing information for data communication.

[0057] (2) The data interface module is associated with the project and equipment management module. It establishes a communication connection based on the communication parameters (local intranet port, local machine number, external server IP address and port number, etc.) configured for the current project in the module. It is responsible for receiving multi-source data from data aggregation nodes or terminal devices and sending analysis results and control instructions to the terminal devices.

[0058] (3) Model training module, including training data management submodule and training execution submodule; wherein, the training data management submodule is used to receive trace images and their label data after manual correction by the handheld acquisition terminal, form a training sample set, and support region selection, direction labeling and positive and negative sample classification of imported sample images; the training execution submodule responds to the training instructions of the management personnel, calls the training sample set to iteratively train and optimize the deep learning model, generates and updates the ignition source trace recognition model, the spread direction trace recognition model and the burn trace recognition model, and pushes the new model to the recognition module; the model training module supports region selection of imported sample images, saves the selected area as a positive sample and the unselected area as a negative sample, saves all sample images in the specified working directory, and supports the establishment of independent training datasets for combustion spread direction, ignition source type and burn trace respectively.

[0059] (4) Trace recognition module, including model loading submodule and image analysis submodule; wherein, the model loading submodule is responsible for loading and managing the latest recognition model generated by the model training module; the image analysis submodule responds to the arrival of a new image, automatically calls the corresponding model in the model loading submodule, analyzes and processes the received trace image, identifies the direction of fire spread, the type of ignition source and the type of burn marks in the image, and outputs the recognition result; the trace recognition module supports remote real-time recognition of images uploaded by handheld devices, and also supports offline recognition of imported local images.

[0060] The traces identified by the trace recognition module include at least: the direction of fire spread determined by the morphology of vegetation burns and the tilt angle of the charred layer; the electric ignition source determined by the characteristics of electrical melting marks; the lightning ignition source determined by the characteristics of lightning melting marks; and the traces of electrical burn-off and lightning burn-off determined by the characteristics of burn-off.

[0061] (5) Spatial analysis and plotting module, used to receive the three-dimensional modeling data of the UAV mapping terminal and overlay and fuse it with the electronic map of the geographic information system. It supports loading three-dimensional model files in 3D tile format and can manually adjust the fusion height parameter between the model and the electronic map. Combined with the recognition results output by the trace recognition module and its corresponding geographic coordinates, it dynamically plots the vector arrow of the fire spread trend and the hot spot area of ​​the ignition source on the electronic map, and supports clicking the plotting icon to view the corresponding original trace photos.

[0062] (6) Task collaboration and push module, running in the background system, is used for:

[0063] When a new image containing suspected signs of spread or ignition sources is received from the handheld acquisition terminal via the data interface module, the geographic coordinate range of the mark is analyzed, a high-priority mapping instruction is generated and sent to the UAV mapping terminal via the data interface module, so that the marked area is given priority for close-up detailed investigation; when the recognition result output by the trace recognition module is received, the spatial analysis and plotting module is triggered to perform update calculations;

[0064] When the spatial analysis and plotting module generates new plotting information, it pushes the information to the display screens of all online handheld acquisition terminals in real time through the data interface module.

[0065] In response to the group chat command from the back-end operators, a voice message is broadcast to all online handheld data acquisition terminals through the data interface module.

[0066] (7) Case management and filing module, which is used to automatically archive all on-site photos, videos, UAV 3D models, recognition results and manual correction records involved in a single fire incident to form a complete electronic case file, and supports quick access and import and export of historical case data;

[0067] (8) Time synchronization engine, used to intelligently associate and align multi-source data of the same fire scene uploaded by different terminals at different times based on timestamps and geographic coordinates, providing a time-consistent data foundation for the spatial analysis and plotting module;

[0068] (9) User permission management module, which supports two permission levels: super administrator and ordinary user. Super administrator has the permission to add, delete and modify users, while ordinary user only has system usage permission and no user management permission.

[0069] II. System Workflow: Workflow diagram as follows Figure 2 As shown.

[0070] 1. On-site modeling and map preparation:

[0071] Before the fire investigation began, aerial mapping of the fire scene was first conducted using a drone mapping terminal. Operators accessed the flight path planning interface via remote control, created a new flight mission and a planar flight path, selected the flight area on the map, and set mapping parameters, including mapping accuracy (e.g., orthophoto GSD) and altitude mode (selecting "relative ground height"). After takeoff, the drone automatically executed the mapping task, recording GPS coordinates and flight altitude in real time during the flight. Once mapping was complete, high-resolution photos from the drone's memory card were copied to a computer. Third-party software such as ContextCapture was used to generate 3D format files (e.g., OSGB format), which were then imported into the spatial analysis and plotting module of the backend system. These files were then overlaid and merged with electronic maps from a geographic information system (e.g., Tianditu) to complete the 3D model of the fire scene.

[0072] 2. On-site investigation and trace collection:

[0073] Survey personnel entered the fire scene carrying handheld data acquisition terminals. Upon discovering suspected signs of fire spread or ignition sources, they used the terminals to take photographs. During the photographing process, the device automatically encapsulated information such as the image, geographic coordinates, shooting angle, and device orientation into a structured data packet.

[0074] When investigators are particularly interested in a particular trace, they can mark it on the image using a handheld data acquisition terminal. This could involve selecting a suspected trace area or adding annotations. The marked image is then uploaded to the backend system via a data aggregation node or a 4G / 5G public network.

[0075] 3. Human-machine collaborative task scheduling:

[0076] When the task coordination and push module of the backend system receives a new image uploaded by the handheld acquisition terminal containing markers indicating possible spread or ignition sources, it analyzes the geographic coordinates of the marker, generates a high-priority mapping instruction, and sends it to the UAV mapping terminal via the data interface module. Upon receiving the instruction, the UAV mapping terminal prioritizes interrupting its current task, automatically generates a detailed planar flight path mapping task covering the marked area, and conducts a close-up, detailed survey of the marked area after takeoff, acquiring high-definition imagery and thermal imaging data. The video stream is then transmitted to the backend system via the real-time image transmission module. Backend operators can observe the images transmitted back by the UAV in real time for further analysis.

[0077] 4. Trace recognition and result push:

[0078] The trace recognition module of the backend system analyzes and processes the received trace images. The recognition process includes: preprocessing the image (denoising, image enhancement, and size normalization), and then calling the latest recognition model (including the ignition source trace recognition model, the spread direction trace recognition model, and the burn trace recognition model) in the model loading submodule to intelligently recognize the image and output the recognition results of the fire spread direction, ignition source type, and burn trace type.

[0079] The traces identified by the trace recognition module include: the direction of fire spread determined by the characteristics of vegetation burn morphology and the tilt angle of the charred layer; the electric ignition source determined by the characteristics of electric melting marks; the lightning ignition source determined by the characteristics of lightning melting marks; and the traces of electric burn-off and lightning burn-off determined by the characteristics of burn-off.

[0080] The identification results are pushed to the handheld data acquisition terminal via the data interface module, where an arrow indicating the direction of fire spread and the predicted range of the ignition source are overlaid on the terminal's display screen. Simultaneously, the identification results are also pushed to the spatial analysis and plotting module.

[0081] 5. Spatial Analysis and Plotting:

[0082] The spatial analysis and plotting module receives 3D modeling data from the UAV mapping terminal and overlays it with the electronic map of the geographic information system. It supports loading 3D model files in 3D Tiles format and allows manual adjustment of the fusion height parameters between the model and the electronic map. Combining the recognition results output by the trace recognition module and their corresponding geographic coordinates, it dynamically plots vector arrows indicating the fire spread trend and hotspot areas of the ignition source on the electronic map. Clicking on the plotted icons allows users to view the corresponding original trace photos.

[0083] When the spatial analysis and plotting module generates new plotting information, the task collaboration and push module pushes this information to the display screens of all online handheld acquisition terminals in real time through the data interface module, achieving multi-terminal synchronous visualization.

[0084] 6. Manual correction and model iteration:

[0085] After viewing the identification results on a handheld data acquisition terminal, surveyors can manually confirm or correct the results based on their professional knowledge.

[0086] For example:

[0087] When the spread direction identified by the system does not match the actual observation, the surveyor can correct it by rotating the direction arrow on the touch screen.

[0088] When the marked ignition source position is off, the marked point can be dragged to correct the position;

[0089] When the system fails to recognize trace features, the investigator can manually add markers.

[0090] The corrected information is returned to the model training module of the backend system via the communication module. The model training module receives the manually corrected trace images and their label data, forming a training sample set. In response to training instructions from management personnel, it uses the training sample set to iteratively train and optimize the deep learning model, generating and updating the ignition source trace recognition model, the spread direction trace recognition model, and the burn trace recognition model, and then pushes the new models to the trace recognition module. Through this design, the system continuously learns and optimizes in actual use, gradually improving its recognition accuracy.

[0091] 7. Case Management and Archiving:

[0092] The case management and filing module of the back-end system automatically archives all on-site photos, videos, drone 3D models, recognition results, and manual correction records related to a single fire incident, forming a complete electronic case file that supports quick access and import / export of historical case data.

[0093] The time-series synchronization engine intelligently associates and aligns multi-source data from the same fire scene uploaded by different terminals at different times based on timestamps and geographic coordinates, providing a time-consistent data foundation for the spatial analysis and plotting modules, and ensuring the uniformity of multi-source data on the time axis and spatial coordinates.

Claims

1. A survey system for post-forest fire investigation and evidence collection, characterized in that, include: Handheld acquisition terminal: used to collect trace images containing geographic coordinates and sensor information on site and send them to the back-end system, while receiving and displaying the analysis results of the back-end system; the handheld acquisition terminal supports switching between intranet direct connection mode and extranet cloud server forwarding mode, and automatically switches to extranet cloud server for data transmission when intranet connection fails; Unmanned aerial vehicle (UAV) mapping terminal: used to conduct aerial mapping of fire scenes to obtain multispectral image data for constructing 3D models of the scene; Data aggregation node: includes a 4G / 5G network router and a wireless power amplifier, used to build a 2.4GHz dedicated wireless network that supports real-time local interaction of multiple terminals in offline or weak network environments, receive and cache data uploaded by the handheld collection terminal, and interact with the backend system; Long-distance data transmission uses a multi-hop network composed of multiple data aggregation nodes to relay data. The backend system includes: Trace recognition module: used to analyze and process the received trace images, identify the direction of fire spread, the type of ignition source and the type of burn marks in the images, and output the recognition results; Spatial analysis and plotting module: used to receive multispectral image data from the UAV mapping terminal and overlay and fuse it with the electronic map of the geographic information system; combined with the recognition results output by the trace recognition module and its corresponding geographic coordinates, dynamically plot the vector arrow of the fire spread trend and the hot spot area of ​​the ignition source on the electronic map; Task collaboration and push module: running in the background system, when a new image containing suspected spread traces or ignition source traces is received from a handheld acquisition terminal, the module analyzes the geographical coordinate range of the marker, generates a high-priority mapping instruction and sends it to the UAV mapping terminal, so that it can prioritize close-up detailed investigation of the marked area. Manual correction module: running on the handheld acquisition terminal, used to receive and overlay the recognition results pushed back by the backend system on the display screen, and allow on-site personnel to manually confirm or correct the recognition results based on their professional knowledge, and return the correction information to the backend system.

2. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, The backend system also includes: The model training module is used to receive trace images and their label data that have been manually corrected by the handheld acquisition terminal, form a training sample set, and call the training sample set in response to the training command to iteratively train and optimize the deep learning model, generate and update the ignition source trace recognition model, the spread direction trace recognition model and the burn trace recognition model, and push the new model to the trace recognition module.

3. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, The handheld data acquisition terminal includes: The multi-sensor fusion imaging module is used to simultaneously trigger the gravity sensor, orientation sensor and global positioning system when collecting trace images, and encapsulate the image, geographic coordinates, shooting angle and device attitude information at the moment of shooting into a structured data packet; The communication module supports uploading structured data packets via a 4G / 5G public network or a dedicated 2.4GHz wireless link established with the data aggregation node, and receiving analysis results from the background system; the communication module automatically displays the intranet or extranet connection status according to the network connection status, and provides photo retransmission function in the event of network outage; The voice intercom module is used to receive or initiate group voice communication with the backend system and other handheld acquisition terminals through the communication module.

4. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, The UAV mapping terminal includes: The autonomous task planning module is used to automatically generate a detailed planar route mapping task covering the target area after receiving the mapping instructions issued by the task collaboration and push module; it has power monitoring and automatic return functions, and can automatically resume unfinished mapping tasks after the battery is replaced. The multispectral remote sensing module integrates a high-definition visible light camera and a thermal infrared imager to simultaneously acquire visible light images and surface thermal radiation distribution data at the fire scene, providing auxiliary data for the identification of hidden fire sources and basic image data for background 3D modeling. The real-time image transmission module supports the real-time transmission of video streams captured by the drone's camera to the back-end system via a real-time streaming protocol, allowing back-end operators to observe and capture images in real time.

5. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, The data aggregation node includes: 4G / 5G network routers are used to install SIM cards to access the public network and provide internet connectivity for aggregation nodes; The AP power amplifier is connected to the 4G / 5G network router via a wired connection. It is used to amplify the 2.4GHz wireless signal, extend the wireless coverage, and support multi-level bridging to further extend the communication distance. The AP power amplifier is configured in relay bridging mode to scan and connect to the upstream wireless network, while providing a local 2.4GHz wireless access point.

6. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, The backend system also includes: The case management and filing module is used to automatically archive all on-site photos, videos, drone 3D models, identification results, and manual correction records related to a single fire incident, forming a complete electronic case file; The time-series synchronization engine is used to intelligently associate and align multi-source data of the same fire scene uploaded by different terminals at different times, based on timestamps and geographic coordinates, providing a time-series consistent data foundation for the spatial analysis and plotting module.

7. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, The traces identified by the trace recognition module include at least: the direction of fire spread determined based on the characteristics of vegetation burn morphology and the tilt angle of the charred layer; the electric ignition source determined based on the characteristics of electric melting marks; the lightning ignition source determined based on the characteristics of lightning melting marks; and the traces of electric burn-off and the traces of burn-off spread determined based on the characteristics of burn-off.

8. The exploration system for post-forest fire investigation and evidence collection according to claim 2, characterized in that, The model training module supports region selection of imported sample images, saving the selected region as a positive sample and the unselected region as a negative sample, and supports the creation of independent training datasets for combustion spread direction, ignition source type and burn marks.

9. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, It also includes a public cloud server, which is configured with a fixed public IP address or bound domain name and has open specified communication ports. It is used as a data relay server to realize remote data communication between the handheld acquisition terminal and the backend system when the handheld acquisition terminal cannot connect to the internal network backend.

10. The exploration system for post-forest fire investigation and evidence collection according to claim 1, characterized in that, Voice communication is supported between multiple handheld terminals and between the handheld terminals and the backend system; the task collaboration and push module is also used to respond to the group chat command of the backend operator and broadcast voice messages to all online handheld acquisition terminals.

Images