Wild animal channel monitoring method based on infrared induction and AI identification

Abstract

The invention relates to the technical field of intelligent monitoring of wild animals, and provides a wild animal channel monitoring method based on infrared induction and AI identification, which comprises the following steps: triggering high-resolution infrared image acquisition when an infrared induction module detects that the change of heat in a channel exceeds a threshold value; an animal target in the image sequence is identified through an AI target detection model, a motion area is extracted, and a transverse / longitudinal motion vector diagram is generated; motion intensity distribution is calculated based on the vector diagram, and a dynamic weight map is generated in combination with a species activity mode database; and carrying out space-time fusion on the dynamic weight map and the infrared images of the current frame and the historical frame, outputting a monitoring result with species classification and an activity track, and uploading the monitoring result to a data center. According to the invention, precise monitoring, intelligent analysis and real-time early warning of wild animal activities can be realized, and the automation level and data value of wild animal channel monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent wildlife monitoring, and more specifically, the present invention relates to a method for monitoring wildlife passages based on infrared sensing and AI recognition. Background Art

[0002] With the continuous improvement of wildlife protection awareness, the monitoring technology of wildlife passages has gradually become an important research direction in the field of ecological protection. Traditional wildlife monitoring methods mainly rely on manual on-site observation and simple camera shooting, and these methods have many limitations. Manual observation not only consumes a large amount of manpower and time, but also is difficult to achieve long-term continuous monitoring and is easily restricted by natural conditions such as weather and terrain. Although traditional camera shooting can record animal activities to a certain extent, it has poor effects at night or under low light conditions and cannot accurately identify animal species and behavior patterns. In recent years, although infrared sensing technology and image recognition technology have been applied to wildlife monitoring to a certain extent, most of these technologies can only achieve simple trigger shooting and preliminary recognition, and cannot conduct in-depth analysis and real-time monitoring of animal behavior. In addition, there are also deficiencies in data transmission and processing in the existing technology. The data transmission efficiency is low, the monitoring results cannot be fed back in time, and there is no effective adaptive update mechanism when facing unidentified species.

[0003] In the process of implementing the embodiments of the present invention, it is found that there are at least the following problems or defects in the existing technology: First, the monitoring effect of the existing monitoring technology is poor at night or under low light conditions, and it is impossible to accurately capture animal activities; Second, the existing image recognition technology does not analyze animal behavior deeply enough and cannot provide detailed movement trajectories and behavior patterns; Third, there are problems of low efficiency and unstable transmission quality in the data transmission process of the existing technology, which cannot meet the requirements of real-time monitoring; Fourth, the existing technology lacks an effective adaptive update mechanism when facing unidentified species and cannot update the model in time to adapt to new monitoring needs. Summary of the Invention

[0004] The present invention provides a method for monitoring wildlife passages based on infrared sensing and AI recognition, including: When the infrared sensing module detects that the heat change in the passage exceeds the first preset threshold, trigger the acquisition of high-resolution infrared images to generate an infrared image sequence of the wildlife passage; Use the AI target detection model to identify animal targets in each frame image of the infrared image sequence, extract the movement areas of the animal targets in each frame image, and generate a horizontal movement vector map and a vertical movement vector map relative to its previous frame image; Calculate the motion intensity distribution of animal targets in each frame of the image based on the horizontal motion vector map and the vertical motion vector map, and generate a dynamic weight map in combination with the species activity pattern database; Perform spatio-temporal fusion on the dynamic weight map and the infrared images of the current frame and historical frames, output the monitoring results with species classification and activity trajectories, and upload them to the data center through a wireless network.

[0005] Further, the triggering of high-resolution infrared image acquisition includes: Real-time monitor the pixel-level heat change of the infrared sensing module. When the heat change amplitude in the same area exceeds the environmental noise threshold in a continuous preset number of frames, activate the infrared camera to take pictures at the highest resolution, and switch to the low-power mode when there is no activity.

[0006] Further, the construction of the AI target detection model includes: Use a lightweight convolutional neural network as the backbone network, add a thermal radiation intensity normalization layer to the infrared image input layer, and linearly map the original thermal radiation value to the standard gray scale interval; When loading the pre-trained species classification model through transfer learning, freeze the parameters of the feature extraction layer except the classification layer; When deploying on edge computing devices, use a model quantization engine to perform low-bitwidth quantization on the network to achieve real-time detection at a preset frame rate, and set the detection threshold to a preset confidence threshold and a preset intersection over union threshold.

[0007] Further, the generation of the horizontal motion vector map and the vertical motion vector map includes: Perform pixel-level matching on the motion regions of the same animal target in two adjacent frames of images, and calculate the horizontal displacement vector and the vertical displacement vector according to the matching results; Map the displacement vectors to the full map according to the spatial positions to generate a horizontal motion vector map and a vertical motion vector map respectively.

[0008] Further, the calculation of the motion intensity distribution includes: According to the displacement amount of each pixel in the horizontal motion vector map, combined with the moving speed threshold of the corresponding species, divide the high-activity area and the low-activity area; Perform the same processing on the vertical motion vector map, and superimpose the horizontal and vertical activity areas to generate a two-dimensional motion intensity distribution map.

[0009] Further, the generation of the dynamic weight map includes: Match the similarity between the motion intensity distribution map and the typical activity templates in the species activity pattern database, and adjust the weight coefficients of each region according to the matching results; Perform time series smoothing processing on the weight coefficients to generate a dynamic weight map for spatio-temporal fusion.

[0010] Furthermore, the spatio-temporal fusion includes: Locally enhance the animal target area of the current frame according to the dynamic weight map, and perform multi-frame mean denoising on the background area using historical frames; Retain the species classification label during the fusion process, and connect the movement trajectories of the same target in consecutive frames into a complete path.

[0011] Furthermore, the method also includes a model adaptive update step: When an unrecognized species is detected for a continuous preset number of frames and the target area ratio exceeds the preset area threshold, automatically trigger the image acquisition device to save the target area and adjacent frame images; After performing background difference and histogram equalization preprocessing on the acquired images, generate pseudo-labels through semi-supervised learning, and update the species classification model using the online knowledge distillation algorithm; Freeze the feature extraction layer during incremental training, only update the parameters of the classification layer, perform model verification tests after training is completed, and deploy the updated version when the accuracy improvement exceeds the preset accuracy threshold.

[0012] Furthermore, the data compression and transmission step includes: Perform inter-frame compression coding on the static background area, and the compression ratio is dynamically adjusted according to the wireless network signal strength; When the animal movement trajectory data is converted into a vector format, each trajectory point contains a timestamp, thermal radiation intensity, and movement direction data, and differential coding is used to compress the direction angle; Give priority to uploading the classification information of the preset protected species, and trigger the emergency transmission mode when an abnormal movement pattern is detected, and improve the compression quality of the key frames to above the preset threshold.

[0013] Furthermore, the abnormal behavior analysis includes: Calculate the turning angle formed by adjacent trajectory points in the movement trajectory. When the cumulative turning angle exceeds the angle threshold within a preset time period and the number of occurrences of a single turning angle greater than the sub-angle threshold reaches the preset number, it is determined as a wandering behavior; For speed mutation detection, use sliding window variance analysis to generate a poaching warning when the speed change exceeds the species average speed threshold within a preset time window and the occurrence frequency reaches the preset frequency; During multi-species trajectory analysis, detect the overlap degree of the activity areas of different species through a spatial topology map. When the overlapping area ratio exceeds the area threshold and the duration exceeds the time threshold, comprehensively determine the habitat conflict level in combination with multi-modal sensor data.

[0014] The above embodiments of the present invention have at least the following beneficial effects: The present invention can achieve precise monitoring and efficient analysis of wildlife passages. Through the combination of the infrared induction module and high-resolution infrared image acquisition, it can clearly capture the activity images of wild animals at night or under low-light conditions, ensuring the continuity and integrity of monitoring. The introduction of the AI target detection model makes the recognition of animal targets more accurate, capable of quickly distinguishing different species and extracting their movement areas, generating detailed motion vector maps, and then calculating the motion intensity distribution, providing rich data for studying animal behavior. The generation of the dynamic weight map and the application of spatio-temporal fusion technology can further enhance the accuracy and reliability of the monitoring results, clearly showing the activity trajectories and behavior patterns of animals, and providing strong support for wildlife protection and ecological research.

[0015] In addition, the present invention can also achieve the adaptive update of the model and the efficient transmission of data. When an unrecognized species is detected, the system can automatically trigger image acquisition and preprocessing, and update the species classification model through semi-supervised learning and online knowledge distillation algorithms to ensure the accuracy and adaptability of the model. The data compression and transmission technology can dynamically adjust the compression ratio according to the wireless network signal strength, give priority to uploading key data, and trigger the emergency transmission mode when an abnormal motion pattern is detected, improving the efficiency and reliability of data transmission, reducing the data transmission cost, and at the same time reducing the occupancy of network bandwidth, ensuring that the monitoring data can be uploaded to the data center in a timely and accurate manner, providing strong support for subsequent analysis and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown by way of example and not limitation, wherein: Figure 1 It is a schematic flowchart of a method for monitoring wildlife passages based on infrared induction and AI recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present invention, and do not limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to convey the scope of the present invention fully to those skilled in the art.

[0018] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, a device, equipment, a method, or a computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0019] It should be noted that the number of any elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0020] The following refers to Figure 1 , Figure 1 which is a schematic flowchart of a wild animal passage monitoring method based on infrared sensing and AI recognition provided for an embodiment of the present invention. As Figure 1 shown, a wild animal passage monitoring method based on infrared sensing and AI recognition includes: S1. When the infrared sensing module detects that the heat change in the passage exceeds a first preset threshold, trigger high-resolution infrared image acquisition to generate an infrared image sequence of the wild animal passage; S2. Use an AI target detection model to perform animal target recognition on each frame image in the infrared image sequence, extract the motion regions of the animal targets in each frame image, and generate a horizontal motion vector map and a vertical motion vector map relative to its previous frame image; S3. Calculate the motion intensity distribution of the animal targets in each frame image based on the horizontal motion vector map and the vertical motion vector map, and combine with the species activity pattern database to generate a dynamic weight map; S4. Perform spatio-temporal fusion on the dynamic weight map and the infrared images of the current frame and historical frames, output a monitoring result with species classification and activity trajectories, and upload it to the data center through a wireless network.

[0021] It should be noted that when the infrared sensing module detects that the heat change in the passage exceeds a first preset threshold, high-resolution infrared image acquisition is triggered to generate an infrared image sequence of the wild animal passage. The infrared sensing module here is a device capable of sensing heat changes. It determines whether a wild animal has passed by detecting the heat change in the passage. The heat change exceeding the first preset threshold means that an animal has entered the monitoring area, thus triggering high-resolution infrared image acquisition. High-resolution infrared image acquisition refers to using an infrared camera to capture images at a high resolution to generate an infrared image sequence containing wild animal information. These image sequences contain the thermal radiation information of the animals, which can clearly show the contours and positions of the animals and provide basic data for subsequent analysis.

[0022] Specifically, the pixel-level heat change of the infrared sensing module refers to the minimum unit of heat change that the module can detect, and this change can be sensed by the module's sensor to detect the heat change at different positions in the channel. The first preset threshold is a set heat change value. When the detected heat change exceeds this value, the system considers that an animal has entered the monitoring area. High-resolution infrared image acquisition refers to using an infrared camera to capture images at high resolution, generating an infrared image sequence containing wildlife information. Here, high resolution refers to the degree of image detail that the camera can capture, usually measured by the number of pixels. For example, a high-resolution infrared camera can capture images of 1920×1080 pixels, which means it can provide clear images of animals for subsequent analysis and identification. The infrared image sequence is a series of continuously captured infrared images, which record the activities of animals in the channel and provide rich data for subsequent analysis.

[0023] Preferably, the infrared sensing module can adopt a thermopile sensor, which can convert heat change into an electrical signal to achieve precise detection of heat change in the channel. The first preset threshold can be adjusted according to the size of the animal and the ambient temperature of the channel. For example, for small animals, a lower threshold can be set, while for large animals, a higher threshold can be set. High-resolution infrared image acquisition can be achieved by setting the resolution parameters of the infrared camera. For example, the resolution of the camera can be set to 1920×1080 pixels to ensure that clear images of animals can be captured. When generating the infrared image sequence, the shooting frequency of the camera can be set. For example, 30 frames of images are captured per second to ensure that the continuous activity process of animals can be recorded.

[0024] In some embodiments, the triggering of high-resolution infrared image acquisition includes: Real-time monitoring of the pixel-level heat change of the infrared sensing module. When the heat change amplitude in the same area exceeds the ambient noise threshold in a continuous preset number of frames, activate the infrared camera to shoot at the highest resolution and switch to the low-power mode when there is no activity.

[0025] It should be noted that the pixel-level heat change of the real-time monitoring infrared sensing module is continuously monitored. When the heat change amplitude in the same area exceeds the environmental noise threshold for a preset number of consecutive frames, the infrared camera is activated to capture images at the highest resolution, and it switches to the low-power mode when there is no activity. The pixel-level heat change here refers to the smallest heat change unit that the infrared sensing module can detect, usually measured in pixels as the basic unit. The environmental noise threshold is a reference value set to distinguish the heat change caused by real animal activities from environmental background noise, such as wind and vegetation movement. The preset number of frames refers to the number of frames in continuous monitoring that need to meet the condition that the heat change exceeds the threshold to ensure that the detected is continuous animal activity rather than accidental interference. The low-power mode is an energy-saving state. When no animal activity is detected, the infrared camera switches to this mode to reduce energy consumption and extend the service life of the device.

[0026] Specifically, the pixel-level heat change of the infrared sensing module is achieved through the sensitivity of the sensor, which can detect the heat change of each pixel point. The environmental noise threshold is usually set according to the background heat fluctuation in the actual environment. For example, in an environment with strong wind, the environmental noise threshold can be set relatively high to avoid false triggering. The preset number of frames can be determined according to the average moving speed of the animal and the frame rate of the camera. For example, if the average moving speed of the animal is fast, the preset number of frames can be set smaller; otherwise, it can be set larger. The low-power mode refers to an energy-saving state that the camera enters when no activity is detected. At this time, the power consumption of the camera will be significantly reduced, usually achieved by reducing the sampling rate of the camera or turning off some functions. For example, when the heat change amplitude in the same area exceeds the environmental noise threshold in 10 consecutive frames, it can be considered that there is animal activity, and the infrared camera is activated to capture images at the highest resolution. When there is no activity, the camera switches to the low-power mode to save energy.

[0027] Preferably, the infrared sensing module can adopt a high-sensitivity thermopile sensor, which can generate measurable electrical signals under tiny heat changes. The environmental noise threshold can be determined by sampling the background heat fluctuation in the monitoring environment for a period of time. For example, by analyzing the background heat changes at different times within a month, a suitable threshold is set. The preset number of frames can be specifically set according to the average moving speed of the animal and the frame rate of the camera. For example, for small and fast-moving animals, the preset number of frames can be set to 5 frames, while for large and slow-moving animals, it can be set to 10 frames. The low-power mode can be achieved by setting the power management parameters of the camera. For example, when no heat change is detected within 30 seconds, the camera automatically switches to the low-power mode. At this time, the power consumption of the camera can be reduced to less than 10% of the normal power consumption, effectively extending the battery life of the device.

[0028] In some embodiments, the construction of the AI object detection model includes: Using a lightweight convolutional neural network as the backbone network, adding a thermal radiation intensity normalization layer to the infrared image input layer, and linearly mapping the original thermal radiation value to the standard grayscale interval; When loading a pre-trained species classification model through transfer learning, freezing the parameters of the feature extraction layer except the classification layer; When deploying on an edge computing device, using a model quantization engine to perform low-bitwidth quantization on the network to achieve real-time detection at a preset frame rate, and setting the detection threshold to a preset confidence threshold and a preset intersection over union threshold.

[0029] It should be noted that the construction of the AI object detection model includes using a lightweight convolutional neural network as the backbone network, adding a thermal radiation intensity normalization layer to the infrared image input layer, and linearly mapping the original thermal radiation value to the standard grayscale interval; when loading a pre-trained species classification model through transfer learning, freezing the parameters of the feature extraction layer except the classification layer; when deploying on an edge computing device, using a model quantization engine to perform low-bitwidth quantization on the network to achieve real-time detection at a preset frame rate, and setting the detection threshold to a preset confidence threshold and a preset intersection over union threshold. Here, the lightweight convolutional neural network refers to an optimized neural network structure designed to reduce the computational amount and model size while maintaining high accuracy. The thermal radiation intensity normalization layer is a network layer used to standardize the thermal radiation value in the infrared image to a fixed grayscale interval for subsequent processing. Transfer learning is a machine learning method that accelerates the training process and improves the model performance by transferring the parameters of a pre-trained model to a new task. The model quantization engine is a tool used to convert model parameters from floating-point numbers to low-bitwidth quantized representations to reduce the storage requirements and computational complexity of the model.

[0030] Specifically, the lightweight convolutional neural network improves the efficiency of the model by reducing the number of parameters and the amount of computation in the network. For example, lightweight architectures such as MobileNet or ShuffleNet can be used. The role of the thermal radiation intensity normalization layer is to linearly map the thermal radiation values in the infrared image from the original range, such as 0 to 1000, to the standard grayscale interval, such as 0 to 255, which can ensure the comparability of image data under different devices or different environments. In transfer learning, the pre-trained species classification model is a model trained on a large-scale dataset. By freezing the parameters of the feature extraction layer and only updating the parameters of the classification layer, it can quickly adapt to the new species classification task. The model quantization engine can quantize the model parameters from 32-bit floating-point numbers to 8-bit integers, thus significantly reducing the model size and the amount of computation. The preset frame rate refers to the number of image frames that the model can process on the edge computing device, such as 30 frames per second. The detection thresholds include the preset confidence threshold and the preset intersection over union (IoU) threshold. The preset confidence threshold is used to determine the confidence of the model in the detection result, usually set between 0.5 and 0.9; the preset IoU threshold is used to evaluate the accuracy of the detection box, usually set to 0.5.

[0031] Preferably, the construction of the lightweight convolutional neural network can include selecting a suitable network architecture, such as MobileNetV2, which reduces the amount of computation through depthwise separable convolutions. The specific implementation of the thermal radiation intensity normalization layer can be to subtract the minimum value from the thermal radiation value of each pixel, divide by the difference between the maximum value and the minimum value, and then multiply by 255. In transfer learning, the pre-trained model can be a ResNet model trained on the ImageNet dataset. By freezing the parameters of the first few layers and only fine-tuning the last classification layer, it can quickly adapt to the new species classification task. The model quantization engine can use tools such as TensorFlow Lite or ONNX Runtime to quantize the model parameters from 32-bit floating-point numbers to 8-bit integers while maintaining the accuracy of the model. When deploying on the edge computing device, the performance of the model can be optimized by adjusting the input size and quantization parameters of the model. For example, the input image size can be adjusted to 224×224 pixels, and the quantization parameters can be set to 8-bit integers. The preset frame rate can be set according to the computing power of the device. For example, for high-performance edge computing devices, it can be set to 60 frames per second, while for low-power devices, it can be set to 15 frames per second. The detection thresholds can be adjusted according to the specific application scenario. For example, in scenarios with high requirements for detection accuracy, the preset confidence threshold can be set to 0.8, and the preset IoU threshold can be set to 0.6.

[0032] In some embodiments, generating the horizontal motion vector map and the vertical motion vector map includes: Perform pixel-level matching on the motion regions of the same animal target in two adjacent frames of images, and calculate the horizontal displacement vector and the vertical displacement vector according to the matching results; Map the displacement vectors to the entire image according to the spatial positions, and generate a horizontal motion vector map and a vertical motion vector map respectively.

[0033] It should be noted that the process of generating the horizontal motion vector map and the vertical motion vector map is to perform pixel-level matching on the motion regions of the same animal target in two adjacent frames of images, calculate the horizontal displacement vector and the vertical displacement vector according to the matching results, and map these displacement vectors to the entire image according to the spatial positions, and generate a horizontal motion vector map and a vertical motion vector map respectively. The pixel-level matching here refers to performing precise matching at the pixel level of the image to determine the position change of the animal target in two adjacent frames of images. The horizontal displacement vector and the vertical displacement vector respectively represent the displacement of the animal target in the horizontal and vertical directions, and these vectors can reflect the motion direction and speed of the animal.

[0034] Specifically, the pixel-level matching is achieved by comparing the gray values or features of each pixel in two adjacent frames of images. For example, the Optical Flow method can be used to calculate the displacement between pixels. The horizontal displacement vector and the vertical displacement vector are obtained by calculating the displacement of each pixel in the horizontal and vertical directions. These displacement vectors can be represented as arrows pointing from a pixel point to its corresponding point in the next frame of the image. The spatial position mapping refers to placing these displacement vectors at the corresponding positions in the image to form a complete vector field. The horizontal motion vector map and the vertical motion vector map respectively represent the motion conditions of the animal target in the horizontal and vertical directions, and these maps can be used for further analysis of the animal's motion pattern.

[0035] Preferably, the pixel-level matching can adopt an improved optical flow method, such as the Farneback optical flow algorithm, which can handle a large range of displacements and has a certain robustness to noise. When calculating the displacement vector, the gray value of each pixel point can be calculated by difference to determine its displacement in the horizontal and vertical directions. For example, for each pixel point, calculate the gray value difference in the horizontal direction in the next frame of the image. If the difference is greater than a certain threshold, it is considered that the pixel has a displacement in the horizontal direction. Similarly, the displacement in the vertical direction can be calculated. When generating the motion vector map, the displacement vector of each pixel can be normalized for subsequent analysis and visualization. For example, the length of the displacement vector can be normalized to the range of 0 to 1 while keeping its direction unchanged. The generated horizontal motion vector map and vertical motion vector map can more intuitively display the motion direction and speed distribution of the animal target, providing accurate data support for subsequent motion intensity analysis.

[0036] In some embodiments, calculating the motion intensity distribution includes: Dividing a high-activity area and a low-activity area according to the displacement amount of each pixel in the horizontal motion vector map and in combination with the movement speed threshold of the corresponding species; Performing the same processing on the vertical motion vector map, and superimposing the horizontal and vertical activity areas to generate a two-dimensional motion intensity distribution map.

[0037] It should be noted that the process of calculating the motion intensity distribution is to divide a high-activity area and a low-activity area according to the displacement amount of each pixel in the horizontal motion vector map and in combination with the movement speed threshold of the corresponding species; performing the same processing on the vertical motion vector map, and superimposing the horizontal and vertical activity areas to generate a two-dimensional motion intensity distribution map. Here, the motion intensity distribution refers to the activity intensity of an animal in different regions of an image. The activity area is divided by the displacement amount and the speed threshold, so as to more intuitively display the activity of the animal. The movement speed threshold is a reference value set according to the average movement speed of the animal, and is used to distinguish the high-activity area and the low-activity area.

[0038] Specifically, the displacement amount of each pixel in the horizontal motion vector map represents the motion intensity of the pixel in the horizontal direction, and the displacement amount of each pixel in the vertical motion vector map represents the motion intensity of the pixel in the vertical direction. The movement speed threshold is set according to the average movement speed of the animal. For example, for a fast-moving animal, the threshold can be set higher, while for a slow-moving animal, the threshold can be set lower. The high-activity area refers to the area where the displacement amount exceeds the movement speed threshold, and the low-activity area refers to the area where the displacement amount is lower than the threshold. By superimposing the horizontal and vertical activity areas, a two-dimensional motion intensity distribution map can be generated, which can intuitively display the activity intensity distribution of the animal in different directions.

[0039] Preferably, the movement speed threshold can be determined by analyzing the motion data of a specific species. For example, by collecting the motion speed data of the species in different environments, calculating its average speed and standard deviation, and then setting the threshold to the average speed plus or minus one standard deviation. When calculating the motion intensity distribution, the displacement amount of each pixel can be normalized to facilitate the comparison of the motion intensity in different directions. For example, the horizontal and vertical displacement amounts are respectively normalized to the range of 0 to 1, and then the normalized displacement amount is compared with the movement speed threshold to divide the activity area. The generated two-dimensional motion intensity distribution map can use color coding to represent different activity intensities. For example, the high-activity area is represented by red, and the low-activity area is represented by blue, so as to more intuitively display the activity pattern of the animal.

[0040] In some embodiments, generating the dynamic weight map includes: Perform similarity matching between the movement intensity distribution map and the typical activity templates in the species activity pattern database, and adjust the weight coefficients of each region according to the matching results; Perform time-series smoothing processing on the weight coefficients to generate a dynamic weight map for spatio-temporal fusion.

[0041] It should be noted that the generation of the dynamic weight map is to perform similarity matching between the movement intensity distribution map and the typical activity templates in the species activity pattern database, adjust the weight coefficients of each region according to the matching results, and perform time-series smoothing processing on the weight coefficients to generate a dynamic weight map for spatio-temporal fusion. Here, the dynamic weight map is a weight distribution map used to enhance key regions in image processing, and it can dynamically adjust the weights according to the activity patterns of animals. Similarity matching refers to determining the adjustment method of the weight coefficients by calculating the similarity between the movement intensity distribution map and the typical activity template. Time-series smoothing processing is a processing method used to reduce the fluctuations of the weight coefficients in the time series and ensure the stability of the weights.

[0042] Specifically, the movement intensity distribution map is generated by analyzing the horizontal and vertical movement vector maps, which reflects the activity intensity of animals in different regions. The species activity pattern database stores the typical activity templates of various animals, and these templates are obtained by analyzing a large amount of animal behavior data. The similarity matching process is achieved by calculating the similarity between the movement intensity distribution map and the typical activity template, and the similarity can be measured by calculating the correlation coefficient or Euclidean distance between the two distributions. The weight coefficients are adjusted according to the similarity matching results. For example, if the activity pattern of a certain region is highly similar to the typical template, the weight coefficient of that region will increase. Time-series smoothing processing can be achieved by the moving average method or the exponential smoothing method to reduce the fluctuations of the weight coefficients in the time series and ensure their stability.

[0043] Preferably, the similarity matching can be achieved by calculating the Pearson correlation coefficient between the movement intensity distribution map and the typical activity template. For example, for a specific species, its typical activity template shows a high activity intensity in the central region and a low activity intensity in the edge region. If the correlation coefficient between the movement intensity distribution map and the template is high, it indicates that the current animal's activity pattern is similar to the typical pattern of this species, and the weight coefficient of the corresponding region can be increased. The time-series smoothing processing can adopt the exponential smoothing method, which can assign higher weights to the most recent data and smooth out short-term fluctuations at the same time. For example, the smoothing parameter can be set to 0.5, so that the update of the weight coefficients will be smoother and avoid drastic fluctuations of the weight coefficients due to short-term activity changes. The dynamic weight map generated in this way can better reflect the activity pattern of animals and provide a more accurate weight distribution for subsequent spatio-temporal fusion.

[0044] In some embodiments, the spatio-temporal fusion includes: Locally enhancing the animal target region of the current frame according to a dynamic weight map, and performing multi-frame mean noise reduction on the background region using historical frames; Preserving the species classification label during the fusion process, and connecting the movement trajectories of the same target in consecutive frames into a complete path.

[0045] It should be noted that spatio-temporal fusion refers to locally enhancing the animal target region of the current frame according to a dynamic weight map, performing multi-frame mean noise reduction on the background region using historical frames, preserving the species classification label during the fusion process, and connecting the movement trajectories of the same target in consecutive frames into a complete path. Here, spatio-temporal fusion is an image processing technique that combines information from both the time and space dimensions to improve the quality and information content of the image. The dynamic weight map is a weight distribution map dynamically generated according to the activity patterns of animals, used to enhance the signal of the target region. Multi-frame mean noise reduction is a method of reducing noise by calculating the average of multiple frames of images. The species classification label refers to the information about the animal species retained during the image processing, used for subsequent analysis and recognition.

[0046] Specifically, the dynamic weight map is dynamically generated according to the activity patterns of animals. It contains the weight values of each pixel, and these weight values reflect the importance of the pixel in the image. During the spatio-temporal fusion process, the animal target region of the current frame is locally enhanced according to the dynamic weight map, which means that the pixel values of the target region are adjusted according to the weight values to highlight the target features. The background region is processed by multi-frame mean noise reduction, that is, taking the average of the corresponding pixels in multiple frames of images to reduce the interference of background noise. The species classification label is the key information retained during the image processing. It identifies the species of animals in the image, and these labels are not lost during the fusion process for subsequent analysis and recognition. The connection of the movement trajectories is achieved by analyzing the position changes of the same target in consecutive frames, and connecting these position points to form a complete movement path.

[0047] Preferably, the weight values of the dynamic weight map can be adjusted according to the activity intensity and position of the animal. For example, if the activity intensity in a certain area is high, the weight value of that area can be set larger. When performing local enhancement, it can be achieved by multiplying the pixel values of the current frame by the corresponding weight values. For multi-frame mean denoising, a certain number of historical frames, such as 5 frames or 10 frames, can be selected to calculate the average value of each pixel to reduce background noise. When connecting the motion trajectories, interpolation methods, such as linear interpolation or spline interpolation, can be used to smooth the trajectory line to ensure the continuity and smoothness of the trajectory. In addition, to improve the accuracy of the trajectory, a distance threshold can be set. When the distance between two consecutive trajectory points exceeds this threshold, the trajectory is considered interrupted and needs to be reconnected. Through these refined steps, spatio-temporal fusion can process image data more effectively and provide more accurate results for wildlife monitoring.

[0048] In some embodiments, the method further includes a model adaptive update step: When an unrecognized species is detected for a continuous preset number of frames and the target area ratio exceeds a preset area threshold, the image acquisition device is automatically triggered to save the target area and adjacent frame images; After performing background difference and histogram equalization preprocessing on the acquired images, pseudo-labels are generated through semi-supervised learning, and an online knowledge distillation algorithm is used to update the species classification model; During incremental training, the feature extraction layer is frozen, and only the parameters of the classification layer are updated. After training is completed, a model verification test is performed. When the accuracy improvement exceeds a preset accuracy threshold, the updated version is deployed.

[0049] It should be noted that the present invention relates to a model adaptive update mechanism. When an unrecognized species is detected for a continuous preset number of frames and the target area ratio exceeds a preset area threshold, the image acquisition device is automatically triggered to save the target area and adjacent frame images. After performing background difference and histogram equalization preprocessing on the acquired images, pseudo-labels are generated through semi-supervised learning, and an online knowledge distillation algorithm is used to update the species classification model. During incremental training, the feature extraction layer is frozen, and only the parameters of the classification layer are updated. After training is completed, a model verification test is performed. When the accuracy improvement exceeds a preset accuracy threshold, the updated version is deployed. Here, the model adaptive update means that the system can automatically identify and adapt to new species. By saving the image data of unrecognized species, preprocessing and model update are performed to improve the accuracy and adaptability of the model. Background difference is an image processing technology used to remove background noise and highlight the target object. Histogram equalization is an image enhancement technology used to adjust the contrast of the image. Semi-supervised learning is a machine learning method that combines a small amount of labeled data and a large amount of unlabeled data for training. The online knowledge distillation algorithm is a model optimization technology that improves the performance and efficiency of the model by transferring the knowledge of a complex model to a simple model.

[0050] Specifically, the continuous preset number of frames means that when the system continuously detects that the number of frames of unrecognized species reaches a set threshold, the update mechanism is triggered. For example, it can be set to 10 consecutive frames. The target area ratio refers to the area ratio of the unrecognized species in the image. When it exceeds the set threshold, such as 10%, it is considered that the species is worthy of further identification. Background difference is to remove background noise by calculating the difference between the current frame and the background model, and the background model can be established by averaging the initial few frames of images. Histogram equalization is to adjust the histogram distribution of the image to make the contrast of the image more uniform. In semi-supervised learning, pseudo-labels are obtained by screening the prediction results of unlabeled data. For example, the prediction results with a confidence level higher than 0.8 are selected as pseudo-labels. In the online knowledge distillation algorithm, the output of the teacher model, the complex model, is used as the training target of the student model, the simple model, to improve the performance of the student model. Incremental training means training on the basis of an existing model and only updating the parameters of the classification layer to reduce the amount of calculation. The preset accuracy threshold refers to the minimum increase in accuracy after the model is updated. For example, it can be set to 2%.

[0051] Preferably, background difference can be achieved by calculating the pixel difference between the current frame and the background model. For example, when the pixel difference exceeds a certain threshold, such as 10 gray levels, it is considered that the pixel belongs to the target object. Histogram equalization can be achieved by calculating the cumulative histogram of the image and mapping it to a new gray level range. For example, the cumulative histogram is mapped to the gray level range of 0 to 255. In semi-supervised learning, the generation of pseudo-labels can be achieved by setting a confidence threshold, such as 0.9, and selecting the prediction results with a confidence level higher than this threshold as pseudo-labels. In the online knowledge distillation algorithm, the teacher model can be a pre-trained deep convolutional neural network, and the student model can be a lightweight network. The student model is trained by minimizing the output difference between the teacher model and the student model. During incremental training, transfer learning techniques can be used to freeze the parameters of the feature extraction layer and only train the classification layer. For example, the Adam optimizer is used and the learning rate is set to 0.001. Model validation testing can be completed by evaluating the accuracy of the model on the validation set. When the accuracy improvement exceeds the preset accuracy threshold, such as 3%, the updated model is deployed to the monitoring system to improve the ability to identify new species.

[0052] In some embodiments, the data compression and transmission step includes: Perform inter-frame compression encoding on the static background area, and the compression ratio is dynamically adjusted according to the wireless network signal strength; When the animal movement trajectory data is converted into a vector format, each trajectory point contains a timestamp, thermal radiation intensity, and movement direction data, and differential encoding is used to compress the direction angle; Prioritize uploading the classification information of preset protected species, trigger the emergency transmission mode when an abnormal motion pattern is detected, and improve the key frame compression quality to above the preset threshold.

[0053] It should be noted that the data compression and transmission steps include using inter-frame compression coding for the static background area, and the compression ratio is dynamically adjusted according to the wireless network signal strength; when the animal movement trajectory data is converted into vector format, each trajectory point includes a timestamp, thermal radiation intensity, and movement direction data, and differential coding is used to compress the direction angle; prioritize uploading the classification information of preset protected species, trigger the emergency transmission mode when an abnormal motion pattern is detected, and improve the key frame compression quality to above the preset threshold. Here, the inter-frame compression coding is a video compression technology that reduces the data volume by comparing the differences between consecutive frames. The wireless network signal strength refers to the quality of the wireless communication link, which affects the rate and reliability of data transmission. The vector format is a data representation method that uses geometric elements such as points and lines to describe the movement trajectory. Differential coding is a data compression method that reduces the data volume by storing the change amount of the data instead of the original data. The emergency transmission mode is a mechanism that improves the data transmission priority when an abnormal situation is detected.

[0054] Specifically, the inter-frame compression coding reduces redundant data by encoding the differences between consecutive frames. For example, if the background area does not change in consecutive frames, only the data of the changed part is transmitted. The wireless network signal strength can be measured by parameters such as the received signal strength indication (RSSI). When the signal strength is low, the compression ratio can be reduced to reduce transmission errors. The trajectory points in the vector format include a timestamp that records the time when the event occurs, thermal radiation intensity that reflects the thermal characteristics of the animal, and movement direction data that indicates the moving direction of the animal. Differential coding compresses the direction angle. For example, if the direction angle change between consecutive trajectory points is small, only the change amount is transmitted. The emergency transmission mode is activated when an abnormal motion pattern is detected. For example, when the animal's movement trajectory shows abnormal behavior, the system will improve the compression quality of the key frames to ensure the accurate transmission of important data.

[0055] Preferably, inter-frame compression encoding can adopt standard video compression algorithms such as H.264, which can effectively reduce the data volume while maintaining the image quality. The dynamic adjustment of the wireless network signal strength can be achieved by monitoring the RSSI value. For example, when the RSSI value is lower than -70 dBm, the compression ratio is reduced from 50% to 30% to ensure the reliability of data transmission. The motion trajectory data in vector format can be stored by defining a structure, where the timestamp can be a millisecond-level timestamp, the thermal radiation intensity can be a normalized value, and the motion direction data can be an angular value. The compression of the direction angle by differential encoding can be achieved by calculating the difference in the direction angles of adjacent trajectory points. For example, if the change in the direction angle is less than 5 degrees, only this difference is transmitted. The emergency transmission mode can be triggered by setting an abnormal behavior threshold. For example, when the sudden change in the movement speed of an animal exceeds 20%, the system will increase the compression quality of key frames to above a preset threshold, such as 90%, to ensure that these key frames can be transmitted with higher quality, thereby providing more accurate data for subsequent analysis.

[0056] In some embodiments, the abnormal behavior analysis includes: Calculating the turning angle formed by adjacent trajectory points in the motion trajectory. When the cumulative turning angle exceeds the angle threshold within a preset time period and the number of occurrences where the single turning angle is greater than the sub-angle threshold reaches the preset number, it is determined as a wandering behavior; The speed mutation detection adopts sliding window variance analysis. When the speed change exceeds the species average speed threshold within a preset time window and the occurrence frequency reaches the preset frequency, a poaching warning is generated; During the multi-species trajectory analysis, the overlap degree of the activity areas of different species is detected through a spatial topology map. When the proportion of the overlapping area exceeds the area threshold and the duration exceeds the time threshold, the habitat conflict level is comprehensively determined in combination with multi-modal sensor data.

[0057] It should be noted that abnormal behavior analysis includes calculating the turning angle formed by adjacent trajectory points in the movement trajectory. When the cumulative turning angle exceeds the angle threshold within a preset time period and the number of occurrences where the single turning angle is greater than the sub-angle threshold reaches the preset number, it is determined as a wandering behavior; speed mutation detection uses sliding window variance analysis. When the speed change exceeds the species average speed threshold within a preset time window and the occurrence frequency reaches the preset frequency, a poaching warning is generated; during multi-species trajectory analysis, the overlap degree of the activity areas of different species is detected through a spatial topology map. When the proportion of the overlapping area exceeds the area threshold and the duration exceeds the time threshold, the habitat conflict level is comprehensively determined in combination with multi-modal sensor data. Here, abnormal behavior analysis refers to identifying possible abnormal behaviors, such as wandering or poaching behaviors, by analyzing the movement trajectories of animals. The turning angle refers to the angle change between adjacent trajectory points in the animal's movement trajectory. Sliding window variance analysis is a statistical method used to detect whether the speed change is abnormal. The spatial topology map is a map used to represent the distribution of the activity areas of different species. Multi-modal sensor data refers to data from multiple sensors, such as sound, temperature, etc., for comprehensive analysis.

[0058] Specifically, the turning angle is determined by calculating the angle between adjacent trajectory points. For example, if the angle between two trajectory points exceeds a certain threshold, such as 30 degrees, it is considered a significant turn. The angle threshold and the sub-angle threshold are set parameters used to distinguish normal turns from abnormal turns. For example, the angle threshold can be set to 180 degrees, and the sub-angle threshold can be set to 45 degrees. The preset time period and the preset number are parameters used to judge the wandering behavior. For example, if the cumulative turning angle exceeds 360 degrees within 10 minutes and the number of occurrences where the single turning angle is greater than 45 degrees reaches 5 times, it is determined as a wandering behavior. In sliding window variance analysis, the size of the sliding window can be set according to the average speed of the species. For example, for animals with a faster speed, the sliding window can be set to 10 seconds; for animals with a slower speed, the sliding window can be set to 30 seconds. The species average speed threshold is set according to the normal movement speed of the animal. For example, if the average speed of a certain animal is 2 meters per second, the speed threshold can be set to 4 meters per second. The preset frequency is a parameter used to judge whether the speed mutation is frequent. For example, if the occurrence frequency of the speed mutation exceeds 3 times per minute, a poaching warning is generated. In the spatial topology map, the overlap degree of the activity areas is determined by calculating the proportion of the overlapping area of the activity areas of different species. For example, if the proportion of the overlapping area exceeds 50% and the duration exceeds 1 hour, the habitat conflict level is comprehensively determined in combination with multi-modal sensor data.

[0059] Preferably, the calculation of the steering angle can be achieved by the method of vector difference, that is, calculating the angle between the direction vectors of adjacent trajectory points. For example, if the direction vectors of two trajectory points are x1, y1 and x2, y2 respectively, the steering angle can be obtained by calculating the angle between these two vectors. The angle threshold and the sub-angle threshold can be adjusted according to the behavioral characteristics of animals. For example, for some animals that turn frequently, the sub-angle threshold can be appropriately increased. In the sliding window variance analysis, the moving average method can be used to calculate the average speed within the sliding window, and then the variance is calculated to detect whether the speed change is abnormal. For example, if the speed variance within the sliding window exceeds a certain threshold, such as 1 square meter per second, the speed change is considered abnormal. When analyzing multi-species trajectories, the spatial topology map can be constructed by Geographic Information System (GIS) technology, and the overlap degree of activity areas can be determined by calculating the ratio of the area of the overlapping area to the total area. For example, if the activity areas of two species are A and B respectively, and the overlapping area is C, the overlap degree can be obtained by calculating the ratio of the area of C to the sum of the areas of A and B. When combining multi-modal sensor data, machine learning algorithms such as decision trees or random forests can be used to comprehensively analyze the data of different sensors, so as to more accurately determine the level of habitat conflict. For example, if the sound sensor detects abnormal sounds and the temperature sensor detects temperature changes at the same time, the level of habitat conflict can be increased.

[0060] The above-mentioned various embodiments of the present invention have the following beneficial effects: The present invention can comprehensively improve the monitoring efficiency of wild animals. Through the infrared induction trigger mechanism, the image acquisition timing can be intelligently controlled, significantly reducing energy consumption while ensuring the monitoring effect; adopting a lightweight network architecture combined with thermal radiation normalization processing can achieve efficient target detection on edge devices; the motion vector analysis technology can accurately capture the animal motion characteristics, and the generation of the dynamic weight map can optimize the monitoring weights of different regions; the spatio-temporal fusion algorithm can improve the image quality and reconstruct the complete activity trajectory; the adaptive model update mechanism can continuously expand the recognition ability of the system; the intelligent data compression strategy can dynamically adjust the transmission parameters according to the network conditions to ensure the priority transmission of key data.

[0061] The present invention can achieve multi-dimensional ecological monitoring functions. Infrared thermal imaging technology can break through the time and environmental limitations of traditional monitoring; the calculation of the motion intensity distribution can quantitatively analyze the animal behavior characteristics; the matching of the species activity database can improve the accuracy of behavior discrimination; the abnormal trajectory detection algorithm can timely detect potential threat behaviors; the multi-species spatial analysis can evaluate the habitat interaction relationship; the semi-supervised learning mechanism can continuously optimize the system performance. By integrating infrared sensing, intelligent analysis and wireless transmission technologies, a complete intelligent monitoring closed-loop system can be constructed to provide all-weather and high-precision technical support for wild animal protection.

[0062] Further, the storage medium of the embodiment of the present application stores program instructions capable of implementing all the above methods. Among them, the program instructions can be stored in the above storage medium in the form of a software product, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.

[0063] The above description is only some preferred embodiments of the present invention and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the embodiments of the present invention.

Claims

1. A monitoring method for wildlife passages based on infrared sensing and AI recognition, characterized in that, It includes the following steps: When the infrared induction module detects that the heat change in the channel exceeds the first preset threshold, it triggers the acquisition of high-resolution infrared images and generates an infrared image sequence of the wildlife channel; Through the AI target detection model, animal target recognition is performed on each frame image in the infrared image sequence, and the motion regions of the animal targets in each frame image are extracted, and a horizontal motion vector map and a vertical motion vector map relative to its previous frame image are generated; Based on the horizontal motion vector map and the vertical motion vector map, the motion intensity distribution of the animal targets in each frame image is calculated, and combined with the species activity pattern database, a dynamic weight map is generated; The dynamic weight map is spatially and temporally fused with the infrared images of the current frame and historical frames, and the monitoring results with species classification and activity trajectories are output and uploaded to the data center through the wireless network.

2. The method according to claim 1, characterized in that, The triggering of the high-resolution infrared image acquisition includes: Real-time monitoring of the pixel-level heat change of the infrared induction module. When the heat change amplitude in the same area exceeds the environmental noise threshold in a continuous preset number of frames, the infrared camera is activated to take pictures at the highest resolution and switches to the low-power mode when there is no activity.

3. The method according to claim 1, wherein The construction of the AI target detection model includes: Using a lightweight convolutional neural network as the backbone network, adding a thermal radiation intensity normalization layer to the infrared image input layer, and linearly mapping the original thermal radiation value to the standard gray scale interval; When loading the pre-trained species classification model through transfer learning, the parameters of the feature extraction layer except the classification layer are frozen; When deploying on the edge computing device, a model quantization engine is used to perform low-bitwidth quantization on the network to achieve real-time detection at a preset frame rate, and the detection threshold is set to the preset confidence threshold and the preset intersection over union threshold.

4. The method according to claim 1, wherein The generation of the horizontal motion vector map and the vertical motion vector map includes: Performing pixel-level matching on the motion regions of the same animal target in two adjacent frame images, and calculating the horizontal displacement vector and the vertical displacement vector according to the matching result; Mapping the displacement vector to the whole image according to the spatial position, and respectively generating a horizontal motion vector map and a vertical motion vector map.

5. The method according to claim 1, wherein The calculation of the motion intensity distribution includes: According to the displacement amount of each pixel in the horizontal motion vector map, combined with the moving speed threshold of the corresponding species, the high-activity area and the low-activity area are divided; Performing the same processing on the vertical motion vector map, and superimposing the horizontal and vertical activity areas to generate a two-dimensional motion intensity distribution map.

6. The method according to claim 5, wherein The generation of the dynamic weight map includes: Performing similarity matching between the motion intensity distribution map and the typical activity templates in the species activity pattern database, and adjusting the weight coefficients of each region according to the matching result; Performing time series smoothing processing on the weight coefficients to generate a dynamic weight map for spatio-temporal fusion.

7. The method according to claim 1, wherein The spatio-temporal fusion includes: Locally enhancing the animal target area of the current frame according to the dynamic weight map, and using the multi-frame mean denoising of the historical frames for the background area; Retaining the species classification label during the fusion process, and connecting the motion trajectories of the same target in consecutive frames into a complete path.

8. The method according to claim 3, wherein The method further includes a model adaptive update step: When an unrecognized species is detected in a continuous preset number of frames and the target area ratio exceeds the preset area threshold, the image acquisition device is automatically triggered to save the target area and the adjacent frame images; After performing background difference and histogram equalization preprocessing on the acquired images, pseudo-labels are generated through semi-supervised learning, and an online knowledge distillation algorithm is used to update the species classification model; During incremental training, the feature extraction layer is frozen, and only the parameters of the classification layer are updated. After training is completed, a model verification test is carried out. When the accuracy improvement exceeds the preset accuracy threshold, the updated version is deployed.

9. The method according to claim 1, wherein The data compression and transmission steps include: For static background regions, inter-frame compression coding is adopted, and the compression ratio is dynamically adjusted according to the wireless network signal strength; When the animal movement trajectory data is converted into a vector format, each trajectory point contains a timestamp, thermal radiation intensity, and movement direction data, and differential coding is used to compress the direction angle; The classification information of the preset protected species is preferentially uploaded. When an abnormal movement pattern is detected, an emergency transmission mode is triggered, and the compression quality of the key frames is improved to above the preset threshold.

10. The method according to claim 1, wherein Abnormal behavior analysis includes: Calculating the turning angle formed by adjacent trajectory points in the movement trajectory. When the cumulative turning angle within a preset time period exceeds the angle threshold and the number of occurrences where the single turning angle is greater than the sub-angle threshold reaches the preset number, it is determined as a wandering behavior; Velocity mutation detection uses sliding window variance analysis. When the velocity change within a preset time window exceeds the species average velocity threshold and the occurrence frequency reaches the preset frequency, a poaching warning is generated; During multi-species trajectory analysis, the overlap degree of the activity areas of different species is detected through a spatial topology map. When the proportion of the overlapping area exceeds the area threshold and the duration exceeds the time threshold, the habitat conflict level is comprehensively determined in combination with multi-modal sensor data.

Images