An ai pollution tracing method and system based on laser radar and unmanned aerial vehicle combination

Abstract

The application provides an AI pollution tracing method based on a laser radar and a UAV, applied to a pollution tracing system, comprising a laser radar device, a cloud server, a UAV and a ground client, wherein the laser radar device, the UAV and the ground client are all connected with the cloud server, and the method comprises the following steps: the laser radar device acquires laser point cloud data of a region to be monitored and coordinate information corresponding to hotspot cloud data in a hotspot map generated by the laser point cloud data; the laser radar device sends the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the laser point cloud data to the cloud server through a network signal; high-definition images and thermal imaging information are intelligently identified through an AI algorithm, whether a co-occurrence target exists in the edge of the atmospheric pollutants in the high-definition images is determined first, the relationship between the atmospheric pollutants and the co-occurrence target is inferred, and whether the thermal forming characteristics of the atmospheric pollutants and the co-occurrence target in the high-definition images conform to common sense is analyzed.

Description

Technical Field

[0001] This invention patent relates to the field of pollution source tracing technology, specifically to an AI-based pollution source tracing method and system based on the combined use of lidar and drones. Background Technology

[0002] Air pollution source tracing refers to the use of various technologies and methods to track and determine the sources, transmission paths, and impact ranges of air pollutants, providing a scientific basis for air pollution control and environmental supervision. Nowadays, air pollution source tracing integrates multi-source information such as satellite remote sensing, ground monitoring, and meteorological data, transforming from planar monitoring to three-dimensional monitoring. It combines technologies such as lidar to achieve comprehensive monitoring of air pollution, and combines AI and machine learning algorithms to achieve automated pollution source identification and tracing.

[0003] Chinese Patent Publication No. CN119165116A discloses an air pollution source tracing system and method based on unmanned aerial vehicles (UAVs), comprising: an airborne air monitoring module for acquiring air data; a ground station system for monitoring the UAV's flight status and sending status information to the UAV control platform, and receiving and storing the air data; the airborne air monitoring module is mounted on the UAV, which controls its flight through the UAV control platform, which controls the UAV in real time based on the status information; a data processing module for preprocessing the air data; and a source tracing analysis module for performing pollution source tracing analysis based on the preprocessed air data to obtain the location of the pollution source. Accurate pollution source location and pollutant diffusion simulation provide a scientific basis for the formulation and implementation of environmental policies, and help to more effectively control and reduce air pollution.

[0004] In the aforementioned technologies, the ground station system can monitor the flight status of the UAV in real time and promptly feed back information to the UAV control platform, ensuring the safety of the flight mission and the continuity of data collection. However, UAVs have a strong singularity when performing visual data collection, relying solely on spatial relationships to infer pollution sources, which lacks logical consistency and limits the accuracy of air pollution source tracing. Therefore, there is an urgent need for an AI pollution source tracing method and system based on the combined use of lidar and UAVs. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-based pollution source tracing method and system based on the combined use of lidar and drones, in order to improve the accuracy of air pollution source tracing.

[0006] According to a first aspect of the present disclosure, a preferred embodiment of the present invention provides an AI-based pollution source tracing method based on the combined use of lidar and drones, applied to a pollution source tracing system, including a lidar device, a cloud server, a drone, and a ground client, wherein the lidar device, drone, and ground client are all connected to the cloud server, and the method includes:

[0007] The lidar device acquires the laser point cloud data of the area to be monitored and the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the laser point cloud data.

[0008] The lidar device sends the coordinate information corresponding to the hotspot cloud data in the hotspot map generated from the lidar point cloud data to the cloud server via network signal;

[0009] The cloud server controls the drone to collect at least two visual information items of the area to be monitored based on the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data. The visual information includes high-definition image information and thermal imaging information.

[0010] The cloud server will at least verify the recognition results of high-definition image information and thermal imaging information, and select the recognition results that match each visual information to transmit to the ground client.

[0011] In one embodiment, the cloud server at least verifies the recognition results of high-definition image information and thermal imaging information, selects the recognition results that match each visual information, and transmits them to the ground client, including:

[0012] Edge feature information is extracted from the high-resolution image information using the underlying logic formulas of YOLOv8 to establish a target set, wherein the target set includes air pollution targets and general targets. The edge feature information extraction steps are as follows:

[0013] First, by introducing more skip connections and split operations, the gradient flow is enhanced and the computation is reduced. The code logic formula is expressed as follows:

[0014]

[0015] The Split operation divides the input feature map into two parts: one part passes directly through multiple Bottleneck layers, and the other part is connected as a Shortcut. Finally, the features are fused through a convolutional layer.

[0016] Multi-scale feature fusion is then achieved by concatenating max pooling results of different scales. The code logic formula is as follows:

[0017] ;

[0018] The ordinary recognition targets along the edges of the atmospheric pollution recognition targets in the high-definition image recognition target set are obtained as high-definition image inference targets. Based on the atmospheric pollution recognition targets and the high-definition image inference target set, a first recognition result set is inferred.

[0019] By fusing the high-definition image information with the thermal imaging information, the thermal imaging information features of any target among the atmospheric pollution identification target and the ordinary identification targets along its edge are obtained.

[0020] If all the thermal imaging information features match the preset features of the atmospheric pollution identification target and the ordinary identification target at its edge, then the first identification result set is directly output.

[0021] If the thermal imaging information features of the atmospheric pollution identification target and its edge ordinary identification target do not match their preset features, then the atmospheric pollution identification target and the high-definition image inferred target are modified, and a second identification result set is inferred based on the new set and output.

[0022] In one embodiment, the lidar device acquires lidar point cloud data of the area to be monitored and coordinate information corresponding to the hotspot cloud data in the hotspot map generated from the lidar point cloud data, including:

[0023] The lidar device performs a 360° scan in at least one plane, with a single scan angle of 2° and a scan radius of 6km. One scan surface is divided into 180 scan intervals to obtain a heat map generated from the lidar point cloud data.

[0024] If the difference between the particle reflection detection frequency detected by the lidar device and the standard frequency of particle reflection is greater than the percentage of the standard frequency of particle reflection, then the reflection area within this scanning interval will be marked as hotspot cloud data in the heat map.

[0025] The distance between the lidar device and the hotspot cloud data is calculated based on the reflection time difference of the hotspot cloud data detected within the scanning interval and the lidar wave velocity.

[0026] The coordinate information of the lidar device and the hotspot cloud data relative to the lidar device are fused with the GPS data of the lidar device to obtain the coordinate information of the hotspot cloud data in the hotspot map generated by the lidar point cloud data.

[0027] According to a second aspect of the present disclosure, this invention provides an AI-based pollution source tracing system based on the combined use of lidar and drones, applied to a pollution source tracing system. The system includes a lidar device, a cloud server, a drone, and a ground client, all of which are connected to the cloud server. The system includes:

[0028] The monitoring module is used by the lidar device to acquire the laser point cloud data of the area to be monitored and the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the laser point cloud data.

[0029] The transmission module is used by the lidar device to send the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the lidar point cloud data to the cloud server via network signal;

[0030] The visual acquisition module is used by the cloud server to control the drone to acquire at least two visual information of the area to be monitored according to the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data. The visual information includes high-definition image information and thermal imaging information.

[0031] The AI ​​recognition module is used by the cloud server to verify the recognition results of high-definition image information and thermal imaging information, and select the recognition results that match each visual information to transmit to the ground client.

[0032] In one embodiment, the AI ​​recognition module includes:

[0033] The first recognition module is used to extract edge feature information from the high-definition image information using the underlying logic formulas of YOLOv8, and establish a set of recognition targets. The set of recognition targets includes air pollution recognition targets and ordinary recognition targets. The edge feature information extraction steps are as follows:

[0034] First, by introducing more skip connections and split operations, the gradient flow is enhanced and the computation is reduced. The code logic formula is expressed as follows:

[0035]

[0036] The Split operation divides the input feature map into two parts: one part passes directly through multiple Bottleneck layers, and the other part is connected as a Shortcut. Finally, the features are fused through a convolutional layer.

[0037] Multi-scale feature fusion is then achieved by concatenating max pooling results of different scales. The code logic formula is as follows:

[0038] ;

[0039] The inference module is used to obtain ordinary recognition targets along the edges of atmospheric pollution recognition targets in the high-definition image recognition target set as high-definition image inference targets, and infer a first recognition result set based on the atmospheric pollution recognition targets and the high-definition image inference target set;

[0040] The second identification module is used to fuse the high-definition image information with thermal imaging information to obtain the thermal imaging information features of any target among the atmospheric pollution identification target and the ordinary identification targets along its edge.

[0041] If all the thermal imaging information features match the preset features of the atmospheric pollution identification target and the ordinary identification target at its edge, the first output module directly outputs the first identification result set.

[0042] The second output module is used to modify the air pollution identification target and the high-definition image inferred target if the thermal imaging information features of the air pollution identification target and its edge ordinary identification target do not match its preset features, and to infer the second identification result set based on the new set and output it.

[0043] In one embodiment, the monitoring module includes:

[0044] The scanning module generates a heat map from the laser point cloud data obtained by the lidar device;

[0045] The locking module is used to mark the reflection area within this scanning interval as hotspot cloud data in the heat map if the difference between the particle reflection detection frequency detected by the lidar device and the standard frequency of particle reflection is greater than the standard threshold.

[0046] The calculation module is used to calculate the distance between the lidar device and the hot spot cloud data based on the reflection time difference of the hot spot cloud data detected within the scanning interval and the lidar wave velocity.

[0047] The matching module is used to fuse the coordinate information of the lidar device and the hotspot cloud data relative to the lidar device with the GPS data of the lidar device to obtain the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the lidar point cloud data.

[0048] As can be seen from the above technical solution, the AI ​​pollution source tracing method and system based on the combined use of lidar and drones provided by this invention patent can include the following beneficial effects: This disclosure uses AI algorithms to intelligently identify high-definition images and thermal imaging information. First, based on whether there are co-occurring targets along the edges of air pollutants in the high-definition image, the relationship between air pollutants and their co-occurring targets is inferred. Then, the thermoforming characteristics of air pollutants and their co-occurring targets in the high-definition image are analyzed to verify whether they conform to common sense, thereby verifying the accuracy of the inferred relationship between air pollutants and their co-occurring targets. This significantly improves the accuracy and reliability of pollution source identification, and is particularly suitable for pollution monitoring in complex industrial areas, urban environments, fire emergency response, agricultural straw burning, industrial park leak detection, complex terrain areas, and at night or under low visibility conditions.

[0049] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this disclosure. Attached Figure Description

[0050] To more clearly illustrate the specific embodiments of this invention, the accompanying drawings used in the description of the specific embodiments or prior art will be briefly introduced below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0051] Figure 1 A flowchart of an AI pollution source tracing method based on the combined use of lidar and drones, provided for this invention patent;

[0052] Figure 2 A flowchart of step S40 in an AI pollution source tracing method based on the combined use of lidar and drones, provided for this invention patent;

[0053] Figure 3 A flowchart of step S10 in an AI pollution source tracing method based on the combined use of lidar and drones provided for this invention patent;

[0054] Figure 4 This invention provides a block diagram of an AI pollution source tracing system based on the combined use of lidar and drones. Detailed Implementation

[0055] The embodiments of the technical solution of this invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of this invention and are therefore intended to limit the scope of protection of this invention.

[0056] Figure 1 This invention provides a flowchart of an AI-based pollution source tracing method using lidar and drones. This embodiment provides an AI-based pollution source tracing method using lidar and drones, applied to a pollution source tracing system, including a lidar device, a cloud server, a drone, and a ground client. The lidar device, drone, and ground client are all connected to the cloud server. Figure 1 As shown, the method includes the following steps S10-S40:

[0057] In step S10, the lidar device acquires the laser point cloud data of the area to be monitored and the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the laser point cloud data;

[0058] In this implementation, the coordinate information of hotspot cloud data can accurately identify the distribution area of ​​pollutants, providing key clues for subsequent pollution source location.

[0059] In step S20, the lidar device sends the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the lidar point cloud data to the cloud server via a network signal;

[0060] In this implementation, by transmitting data to the cloud server in real time, the system can respond quickly to pollution events and promptly notify relevant departments to take measures. The cloud server plans the flight path of the drone based on the coordinate information of the hotspot cloud data to ensure that the drone can cover all hotspot areas. The path planning needs to take into account the drone's flight speed, endurance, and the time requirements of the data collection task.

[0061] In step S30, the cloud server controls the drone to collect at least two visual information of the area to be monitored according to the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data. The visual information includes high-definition image information and thermal imaging information.

[0062] In this implementation, drones are characterized by rapid response and flexible maneuverability, enabling them to quickly reach hotspot areas to collect high-definition images and thermal imaging information. At the same time, real-time data transmission ensures that the cloud server can obtain the latest monitoring information in a timely manner. The combination of high-definition images and thermal imaging information provides multi-dimensional data support for pollution source tracing. High-definition images can provide visual details, while thermal imaging information can reveal temperature anomalies. The combination of the two can more comprehensively identify pollution sources.

[0063] In step S40, the cloud server at least checks the recognition results of the high-definition image information and the thermal imaging information, selects the recognition results that match each visual information, and transmits them to the ground client.

[0064] In this implementation, multiple recognition results are evaluated using AI algorithms, and the result that is most logical and reasonable is selected. Based on this, multispectral imaging technology can be used to generate images by capturing spectral information from multiple specific bands. It can not only acquire information in the visible light range, but also extend to near-infrared, short-wave infrared, and other bands. By utilizing the different reflectivities of different substances in different bands, multispectral imaging can more accurately identify target substances and can be used to analyze pollutant composition, concentration, etc.

[0065] In one embodiment, such as Figure 2 As shown, in step S40, the cloud server at least verifies the recognition results of the high-definition image information and the thermal imaging information, selects the recognition results that match each visual information, and transmits them to the ground client, including the following steps S41-S45:

[0066] In step S41, edge feature information in the high-definition image information is extracted using the underlying logic formula of YOLOv8 to establish a target set for identification. This target set includes air pollution targets and general targets. The edge feature information extraction steps are as follows:

[0067] First, by introducing more skip connections and split operations, the gradient flow is enhanced and the computation is reduced. The code logic formula is expressed as follows:

[0068]

[0069] The Split operation divides the input feature map into two parts: one part passes directly through multiple Bottleneck layers, and the other part is connected as a Shortcut. Finally, the features are fused through a convolutional layer.

[0070] Multi-scale feature fusion is then achieved by concatenating max pooling results of different scales. The code logic formula is as follows:

[0071] ;

[0072] In this implementation, edge features refer to areas in the image where brightness or color changes significantly. These typically correspond to the outline of an object or the boundary between different objects. Through edge detection algorithms, the outline of pollutants or objects can be clearly extracted. Based on the outline of pollutants or objects, information such as the type, size, and location of the target can be determined.

[0073] In step S42, ordinary recognition targets along the edges of air pollution recognition targets in the high-definition image recognition target set are obtained as high-definition image inference targets, and a first recognition result set is inferred based on the air pollution recognition targets and the high-definition image inference target set;

[0074] In this implementation method, common air pollution targets include: bare soil, dust, smoke, straw burning, etc. Common ordinary identification targets include: fog cannons, fog cannon machines, fog cannon trucks, soil haulers, non-road vehicles, dust nets, sprinklers, concrete, semi-trailers, moving cars, moving buses, etc. Ordinary identification targets around the air pollution targets are identified as high-definition image inference targets. Based on this, the absence of dust nets near bare soil, the absence of sprinklers or fog cannons near dust, and the presence of moving cars or moving buses near dust are all considered air pollution identification results. It is worth noting that the identification results need to rely on the three-dimensional spatial relationship between the air pollution targets and the high-definition image inference targets to determine the results, that is, at least two perspectives, top view and side view, are required, which can significantly reduce the misjudgment caused by a single perspective.

[0075] In step S43, the high-definition image information is fused with the thermal imaging information to obtain the thermal imaging information features of any target among the atmospheric pollution identification target and the ordinary identification targets along its edge;

[0076] In this implementation, high-definition image information and thermal imaging information can complement each other from the same perspective, reducing the limitations of a single information source and enabling a more comprehensive analysis of the target's characteristics, thereby improving the accuracy of identification. For example, common air pollution targets are dust, and high-definition images can be used to infer targets such as chimneys and moving cars. By analogy with the positions of chimneys and moving cars in the high-definition image information, the temperature distribution of chimneys and moving cars can be obtained from the thermal imaging information.

[0077] In step S44, if all the thermal imaging information features match the preset features of the atmospheric pollution identification target and the ordinary identification target at its edge, the first identification result set is directly output.

[0078] In this implementation, based on step S43, if the temperatures of dust, chimneys, and moving cars are all in a non-high-temperature state in the thermal imaging information, the air pollution identification result is that there is a moving car near the dust (the non-high-temperature chimneys and dust have no logical relationship).

[0079] In step S45, if the thermal imaging information features of the atmospheric pollution identification target and its edge ordinary identification target do not match its preset features, the atmospheric pollution identification target and the high-definition image inferred target are modified, and a second identification result set is inferred based on the new set and output.

[0080] In this implementation, based on step S43, if the temperature of dust and chimney is high in the thermal imaging information, but the temperature of the car is not high, it indicates that the air pollution target is misjudged. The air pollution identification target is changed to smoke, and the air pollution identification result is that there is a chimney near the smoke (the car in a non-high temperature state has no logical relationship with the smoke).

[0081] In one embodiment, such as Figure 3 As shown, in step S10, the lidar device acquires the lidar point cloud data of the area to be monitored and the coordinate information corresponding to the hotspot cloud data in the hotspot map generated from the lidar point cloud data, including the following steps S11-S14:

[0082] In step S11, the lidar device is used to obtain a heat map generated from the lidar point cloud data;

[0083] In this implementation, lidar is a technology that uses lasers for distance measurement and target detection. When running, it performs a 360° scan in at least one plane, with a single scan angle of 2° and a scan radius of 6km. One scan surface is divided into 180 scan intervals. The 360° scan can cover the entire horizontal direction and is performed in at least one plane. It may involve multiple vertical planes to obtain three-dimensional information.

[0084] In step S12, if the difference between the particle reflection detection frequency detected by the lidar device and the standard frequency of particle reflection is greater than the percentage of the standard frequency of particle reflection, then the reflection area within this scanning interval will be marked as hotspot cloud data in the hotspot map.

[0085] In this implementation, when the percentage difference of the reflective area exceeds the standard threshold, the area will be marked as hotspot cloud data in the heat map. Hotspot cloud data indicates that the area may be an area with a high concentration of pollutants and requires further attention and analysis.

[0086] In step S13, the distance between the lidar device and the hot spot cloud data is calculated based on the reflection time difference of the hot spot cloud data detected in the scanning interval and the lidar wave velocity.

[0087] In this implementation, the lidar device emits a laser pulse toward the target area. After the laser pulse encounters the target object, it is reflected back to the lidar device. The device measures the distance to the target object by receiving the reflected light. By measuring the time of flight (TOF) or phase difference of the laser pulse, the distance to the target object is calculated, and three-dimensional hotspot cloud data is generated by combining the scanning angle.

[0088] In step S14, the coordinate information of the lidar device and the hotspot cloud data relative to the lidar device is fused with the GPS data of the lidar device to obtain the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the lidar point cloud data.

[0089] In this implementation, fusing the relative coordinates of the hotspot cloud data acquired by the lidar device with the GPS data from the lidar device is a crucial step in achieving precise positioning of the hotspot cloud data. This process ensures that the coordinates of the hotspot cloud data are converted from relative coordinates to absolute geographic coordinates (such as latitude and longitude), thereby enabling accurate location of pollution sources in the actual environment.

[0090] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein.

[0091] Figure 4 A block diagram of an AI pollution source tracing system based on the combined use of lidar and drones, provided for this invention patent, is shown below. Figure 4 As shown:

[0092] An AI-based pollution source tracing system based on the combined use of lidar and drones is applied to a pollution source tracing system. It includes lidar equipment, a cloud server, drones, and a ground client. The lidar equipment, drones, and ground client are all connected to the cloud server. The system includes:

[0093] The monitoring module 100 is used by the lidar device to acquire the laser point cloud data of the area to be monitored and the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data.

[0094] The transmission module 200 is used by the lidar device to send the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the lidar point cloud data to the cloud server via a network signal;

[0095] The visual acquisition module 300 is used by the cloud server to control the drone to acquire at least two visual information of the area to be monitored according to the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data. The visual information includes high-definition image information and thermal imaging information.

[0096] The AI ​​recognition module 400 is used by the cloud server to verify the recognition results of at least the high-definition image information and the thermal imaging information, and to select the recognition results that match each visual information and transmit them to the ground client.

[0097] In one embodiment, such as Figure 4 As shown, the AI ​​recognition module 400 includes:

[0098] The first recognition module 401 is used to extract edge feature information from the high-definition image information using the underlying logic formula of YOLOv8, and establish a set of recognition targets. The set of recognition targets includes air pollution recognition targets and ordinary recognition targets. The edge feature information extraction steps are as follows:

[0099] First, by introducing more skip connections and split operations, the gradient flow is enhanced and the computation is reduced. The code logic formula is expressed as follows:

[0100]

[0101] The Split operation divides the input feature map into two parts: one part passes directly through multiple Bottleneck layers, and the other part is connected as a Shortcut. Finally, the features are fused through a convolutional layer.

[0102] Multi-scale feature fusion is then achieved by concatenating max pooling results of different scales. The code logic formula is as follows:

[0103] ;

[0104] Inference module 402 is used to obtain ordinary recognition targets along the edges of atmospheric pollution recognition targets in the high-definition image recognition target set as high-definition image inference targets, and infer a first recognition result set based on the atmospheric pollution recognition targets and the high-definition image inference target set;

[0105] The second identification module 403 is used to fuse the high-definition image information with thermal imaging information to obtain the thermal imaging information features of any target among the atmospheric pollution identification target and the ordinary identification targets along its edge.

[0106] If all the thermal imaging information features match the preset features of the atmospheric pollution identification target and the ordinary identification target at its edge, the first output module 404 directly outputs the first identification result set.

[0107] The second output module 405 is used to modify the air pollution identification target and the high-definition image inferred target if the thermal imaging information features of the air pollution identification target and the ordinary identification target along its edge do not match its preset features, and to infer a second identification result set based on the new set and output it.

[0108] In one embodiment, such as Figure 4 As shown, the monitoring module 100 includes:

[0109] The scanning module 101 generates a heat map from the laser point cloud data obtained by the lidar device;

[0110] The locking module 102 is used to mark the reflection area in the heat map as hotspot cloud data if the difference between the particle reflection detection frequency detected by the lidar device and the standard frequency of particle reflection is greater than the standard threshold.

[0111] The calculation module 103 is used to calculate the distance between the lidar device and the hot spot cloud data based on the reflection time difference of the hot spot cloud data detected in the scanning interval and the lidar wave velocity.

[0112] The matching module 104 is used to fuse the coordinate information of the lidar device and the hotspot cloud data relative to the lidar device with the GPS data of the lidar device to obtain the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the lidar point cloud data.

[0113] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0114] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An AI-based pollution source tracing method based on the combined use of lidar and drones, applied to a pollution source tracing system, comprising lidar equipment, a cloud server, drones, and a ground client, wherein the lidar equipment, drones, and ground client are all connected to the cloud server, characterized in that, The method includes: The lidar device acquires the laser point cloud data of the area to be monitored and the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the laser point cloud data. The lidar device sends the coordinate information corresponding to the hotspot cloud data in the hotspot map generated from the lidar point cloud data to the cloud server via network signal; The cloud server controls the drone to collect at least two visual information items of the area to be monitored based on the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data. The visual information includes high-definition image information and thermal imaging information. The cloud server will at least verify the recognition results of high-definition image information and thermal imaging information, and select the recognition results that match each visual information to transmit to the ground client. The cloud server will at least verify the recognition results of high-definition image information and thermal imaging information, select the recognition results that match each visual information, and transmit them to the ground client, including: Edge feature information is extracted from the high-resolution image information using the underlying logic formulas of YOLOv8 to establish a target set, wherein the target set includes air pollution targets and general targets. The edge feature information extraction steps are as follows: First, by introducing more skip connections and split operations, the gradient flow is enhanced and the computation is reduced. The code logic formula is expressed as follows: The Split operation divides the input feature map into two parts: one part passes directly through multiple Bottleneck layers, and the other part is connected as a Shortcut. Finally, the features are fused through a convolutional layer. Multi-scale feature fusion is then achieved by concatenating max pooling results of different scales. The code logic formula is as follows: ； The ordinary recognition targets along the edges of the atmospheric pollution recognition targets in the high-definition image recognition target set are obtained as high-definition image inference targets. Based on the atmospheric pollution recognition targets and the high-definition image inference target set, a first recognition result set is inferred. By fusing the high-definition image information with the thermal imaging information, the thermal imaging information features of any target among the atmospheric pollution identification target and the ordinary identification targets along its edge are obtained. If all the thermal imaging information features match the preset features of the atmospheric pollution identification target and the ordinary identification target at its edge, then the first identification result set is directly output. If the thermal imaging information features of the atmospheric pollution identification target and its edge ordinary identification target do not match its preset features, then modify the atmospheric pollution identification target and the high-definition image inferred target, and infer a second identification result set based on the new set and output it. The lidar device acquires laser point cloud data of the area to be monitored and coordinate information corresponding to the hotspot cloud data in the heat map generated from the laser point cloud data, including: The lidar device is used to generate a heat map from the lidar point cloud data; If the difference between the particle reflection detection frequency detected by the lidar device and the standard frequency of particle reflection is greater than the percentage of the standard frequency of particle reflection, then the reflection area within this scanning interval will be marked as hotspot cloud data in the heat map. The distance between the lidar device and the hotspot cloud data is calculated based on the reflection time difference of the hotspot cloud data detected within the scanning interval and the lidar wave velocity. The coordinate information of the lidar device and the hotspot cloud data relative to the lidar device are fused with the GPS data of the lidar device to obtain the coordinate information of the hotspot cloud data in the hotspot map generated by the lidar point cloud data.

2. An AI-based pollution source tracing system based on the combined use of lidar and drones, characterized in that, An application in a pollution source tracing system includes lidar equipment, a cloud server, drones, and ground clients. The lidar equipment, drones, and ground clients are all connected to the cloud server. The system includes: The monitoring module is used by the lidar device to acquire the laser point cloud data of the area to be monitored and the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the laser point cloud data. The transmission module is used by the lidar device to send the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the lidar point cloud data to the cloud server via network signal; The visual acquisition module is used by the cloud server to control the drone to acquire at least two visual information of the area to be monitored according to the coordinate information corresponding to the hot spot cloud data in the hot spot map generated by the laser point cloud data. The visual information includes high-definition image information and thermal imaging information. The AI ​​recognition module is used by the cloud server to verify the recognition results of high-definition image information and thermal imaging information, and select the recognition results that match each visual information to transmit to the ground client. The AI ​​recognition module includes: The first recognition module is used to extract edge feature information from the high-definition image information using the underlying logic formulas of YOLOv8, and establish a set of recognition targets. The set of recognition targets includes air pollution recognition targets and ordinary recognition targets. The edge feature information extraction steps are as follows: First, by introducing more skip connections and split operations, the gradient flow is enhanced and the computation is reduced. The code logic formula is expressed as follows: The Split operation divides the input feature map into two parts: one part passes directly through multiple Bottleneck layers, and the other part is connected as a Shortcut. Finally, the features are fused through a convolutional layer. Multi-scale feature fusion is then achieved by concatenating max pooling results of different scales. The code logic formula is as follows: ； The inference module is used to obtain ordinary recognition targets along the edges of atmospheric pollution recognition targets in the high-definition image recognition target set as high-definition image inference targets, and infer a first recognition result set based on the atmospheric pollution recognition targets and the high-definition image inference target set; The second identification module is used to fuse the high-definition image information with thermal imaging information to obtain the thermal imaging information features of any target among the atmospheric pollution identification target and the ordinary identification targets along its edge. If all the thermal imaging information features match the preset features of the atmospheric pollution identification target and the ordinary identification target at its edge, the first output module directly outputs the first identification result set. The second output module is used to modify the air pollution identification target and the high-definition image inferred target if the thermal imaging information features of the air pollution identification target and its edge ordinary identification target do not match its preset features, and to infer the second identification result set based on the new set and output it. The monitoring module includes: The scanning module generates a heat map from the laser point cloud data obtained by the lidar device; The locking module is used to mark the reflection area within this scanning interval as hotspot cloud data in the heat map if the difference between the particle reflection detection frequency detected by the lidar device and the standard frequency of particle reflection is greater than the standard threshold. The calculation module is used to calculate the distance between the lidar device and the hot spot cloud data based on the reflection time difference of the hot spot cloud data detected within the scanning interval and the lidar wave velocity. The matching module is used to fuse the coordinate information of the lidar device and the hotspot cloud data relative to the lidar device with the GPS data of the lidar device to obtain the coordinate information corresponding to the hotspot cloud data in the hotspot map generated by the lidar point cloud data.

Images