Intelligent locust population density monitoring method based on AI visual identification

Abstract

The invention relates to the technical field of agricultural pest monitoring, in particular to an intelligent locust population density monitoring method based on AI visual identification, which comprises the following steps: acquiring a locust moving image sequence of a target area through a multispectral camera device, and synchronously acquiring geographic coordinates and environmental parameters; performing moving target detection on the image sequence, and fusing a background difference method and an optical flow method to separate a dynamic locust target; using an improved YOLOv8 deep learning model to identify and classify locust individuals, and outputting target position and category information; performing statistical modeling on the number of the locusts in the high-density overlapping region based on a density map regression network; a three-dimensional population density thermodynamic diagram is generated in combination with a geographic information system (GIS), and a migration path is predicted. In the aspect of detection precision, the absolute error can be less than or equal to 15% within the density range of 100-500 pieces / m < 2 >; in the aspect of data processing efficiency, the processing time for uploading a 1080P image is less than or equal to 0.2 second / frame; and the operation and maintenance cost is reduced by about 70% compared with manual inspection in the aspect of system cost.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural pest monitoring, and specifically to an intelligent monitoring method for locust population density based on AI visual recognition. Background Art

[0002] Locust monitoring and forecasting has long been considered a key task for grassland management and plant protection departments. Timely and effective locust control is a crucial task. However, my country's vast grasslands, primarily distributed in northwest and north China, and in the mountainous areas of southern China, present inconvenient transportation and a severe shortage of control personnel and equipment. This has led to a significant degree of blindness and passivity in locust control efforts.

[0003] Currently, there are no particularly effective methods for monitoring locust populations and locust infestations. Limited reports have been found in domestic and international literature, particularly regarding monitoring and forecasting methods for grassland locusts. Existing technologies primarily rely on ground-based grassland observation stations to observe population density. Forecasting locust infestations relies on human judgment based on changes in locust population density over time, or on remote sensing and image recognition-based monitoring technologies.

[0004] However, the existing monitoring methods have the following problems:

[0005] (1) Manual survey at fixed observation stations: This method relies on visual counting by personnel, which is inefficient (≤2 hectares per person per day) and cannot cope with large-scale locust swarms.

[0006] (2) Remote sensing monitoring: The resolution of satellite images is insufficient (meter level), making it difficult to identify single insect targets.

[0007] (3) Basic image recognition: The conventional CNN model has a false detection rate of >40% in scenarios with overlapping targets and environmental interference.

[0008] In view of this, we propose an intelligent monitoring method for locust population density based on AI visual recognition. Summary of the Invention

[0009] The purpose of the present invention is to provide an intelligent monitoring method for locust population density based on AI visual recognition to solve the problems raised in the above background technology.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] An intelligent monitoring method for locust population density based on AI visual recognition includes the following steps:

[0012] (S1) collecting a sequence of locust activity images in a target area by a multispectral camera device, and synchronously obtaining geographic coordinates and environmental parameters;

[0013] (S2) performing moving target detection on the image sequence, fusing background difference method and optical flow method to separate dynamic locust targets;

[0014] (S3) using the improved YOLOv8 deep learning model to identify and classify locust individuals, and output target location and category information;

[0015] (S4) Statistical modeling of locust populations in high-density overlapping areas based on density map regression networks;

[0016] (S5) Combined with geographic information system (GIS), three-dimensional population density heat maps were generated and migration paths were predicted.

[0017] Preferably, in step S1: the multispectral imaging device includes a visible light band, a near infrared band and a polarized light imaging module;

[0018] Image acquisition is completed through the collaboration of drones and ground fixed monitoring stations.

[0019] Preferably, the step S2 specifically includes:

[0020] Build a dynamic background model based on the ViBe algorithm to separate the foreground moving targets;

[0021] The pyramid Lucas-Kanade optical flow method is used to calculate the locust motion vector;

[0022] The motion trajectory is aggregated and analyzed using the DBSCAN clustering algorithm to eliminate noise interference.

[0023] Preferably, the improved YOLOv8 model includes the following optimizations:

[0024] The backbone network is replaced with the ConvNeXt-Tiny structure to improve the detection capability of small objects;

[0025] The anchor frame parameters are optimized based on locust morphological data, and three scales are set: 8×8, 16×16, and 32×32;

[0026] The detection head integrates the CBAM attention mechanism, and the weight calculation formula is:

[0027] σ(W1·ReLU(W0·Favg)+W1·ReLU(W0·Fmax))

[0028] Among them: Favg and Fmax are average pooling and maximum pooling features respectively.

[0029] Preferably, the density map regression network adopts:

[0030] Lightweight CSRNet architecture, consisting of 10 dilated convolutional layers with a dilation rate of 2;

[0031] Adaptive spatial transformation module, the formula is defined as:

[0032] Used to correct perspective distortion;

[0033] Multi-scale feature fusion branch, the number of output channels is 256.

[0034] Preferably, the GIS system implementation in step S5 includes:

[0035] Generate density distribution surface based on Kriging interpolation method;

[0036] Combining wind speed and direction data to build a migration and diffusion model, the differential equation is:

[0037]

[0038] Where: D is the diffusion coefficient, v is the migration velocity, and λ is the population growth rate;

[0039] The disaster levels are divided according to density thresholds: Level 1 (<10 / ㎡), Level 2 (10-50 / ㎡), and Level 3 (>50 / ㎡).

[0040] Preferably, it also includes:

[0041] The edge computing node executes steps S2-S4 and uses TensorRT to accelerate inference;

[0042] Transmit density data back to the cloud analysis platform via the 5G NR network;

[0043] The monitoring records are encrypted with SHA-256 hash and written into the blockchain for evidence storage.

[0044] Preferably, the locust classification includes:

[0045] Morphological distinction between adults and nymphs;

[0046] The confidence threshold of suspected targets is set to 0.7;

[0047] Mating behavior recognition is achieved through a posture estimation model.

[0048] The present invention further provides an electronic device, comprising:

[0049] Multispectral image acquisition module;

[0050] A processor configured to execute the above method;

[0051] The memory stores a locust feature database and model parameters.

[0052] The present invention also provides a computer-readable storage medium storing program instructions, which implement the above method when executed by a processor.

[0053] Compared with the existing technology, the beneficial effects of the present invention are: in terms of detection accuracy, the intelligent monitoring method of locust population density based on AI visual recognition can achieve an absolute error of ≤15% within the density range of 100-500 locusts / ㎡; in terms of data processing efficiency: the processing time for uploading 1080P images is ≤0.2 seconds / frame; in terms of system cost: the operation and maintenance costs are reduced by about 70% compared with manual inspections. DETAILED DESCRIPTION

[0054] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] An intelligent monitoring method for locust population density based on AI visual recognition includes the following steps:

[0056] (S1) collecting a sequence of locust activity images in a target area by a multispectral camera device, and synchronously obtaining geographic coordinates and environmental parameters;

[0057] (S2) performing moving target detection on the image sequence, fusing background difference method and optical flow method to separate dynamic locust targets;

[0058] (S3) using the improved YOLOv8 deep learning model to identify and classify locust individuals, and output target location and category information;

[0059] (S4) Statistical modeling of locust populations in high-density overlapping areas based on density map regression networks;

[0060] (S5) Combined with geographic information system (GIS), three-dimensional population density heat maps were generated and migration paths were predicted.

[0061] Specifically, a multispectral camera device is used to collect a sequence of locust activity images in the target area, and the geographic coordinates and environmental parameters are obtained simultaneously: the multispectral camera device includes a visible light band (400-700nm), a near-infrared band (700-1100nm) and a polarized light imaging module; image acquisition is completed by a drone (flight altitude 30-100m, resolution 0.5cm / pixel) in collaboration with a ground fixed monitoring station.

[0062] Specifically, performing moving target detection on the image sequence and integrating the background difference method with the optical flow method to separate the dynamic locust targets specifically includes:

[0063] S21: Build a dynamic background model based on the ViBe algorithm to separate the foreground moving targets;

[0064] S22: Apply the pyramid Lucas-Kanade optical flow method to calculate the locust motion vector;

[0065] S23: Use the DBSCAN clustering algorithm to perform cluster analysis on the motion trajectory and eliminate noise interference.

[0066] Specifically, the improved YOLOv8 model includes the following optimizations:

[0067] The backbone network is replaced with the ConvNeXt-Tiny structure to improve the detection capability of small objects;

[0068] The anchor frame parameters are optimized based on locust morphological data, and three scales are set: 8×8, 16×16, and 32×32;

[0069] The detection head integrates the CBAM attention mechanism, and the weight calculation formula is:

[0070] σ(W1·ReLU(W0·Favg)+W1·ReLU(W0·Fmax))

[0071] Among them: Favg and Fmax are average pooling and maximum pooling features respectively.

[0072] Specifically, the density map regression network adopts:

[0073] Lightweight CSRNet architecture, consisting of 10 dilated convolutional layers with a dilation rate of 2;

[0074] Adaptive spatial transformation module, the formula is defined as:

[0075] Used to correct perspective distortion;

[0076] Multi-scale feature fusion branch, the number of output channels is 256.

[0077] Specifically, the generation of a three-dimensional population density heat map in conjunction with a geographic information system (GIS) and the prediction of migration paths by the GIS system include:

[0078] Generate density distribution surface based on Kriging interpolation method;

[0079] Combining wind speed and direction data to build a migration and diffusion model, the differential equation is:

[0080]

[0081] Where: D is the diffusion coefficient, v is the migration velocity, and λ is the population growth rate;

[0082] The disaster levels are divided according to density thresholds: Level 1 (<10 / ㎡), Level 2 (10-50 / ㎡), and Level 3 (>50 / ㎡).

[0083] Specifically, the present invention also includes:

[0084] The edge computing node executes steps S2-S4 and uses TensorRT to accelerate inference;

[0085] Transmit density data back to the cloud analysis platform via the 5G NR network;

[0086] The monitoring records are encrypted with SHA-256 hash and written into the blockchain for evidence storage.

[0087] In this embodiment, the locust classification includes:

[0088] Morphological distinction between adults and nymphs (wing length / body length ratio ≥ 0.6 for adults);

[0089] The confidence threshold of suspected targets is set to 0.7;

[0090] Mating behavior recognition is achieved through a posture estimation model.

[0091] The present invention further provides an electronic device, comprising:

[0092] Multispectral image acquisition module;

[0093] A processor configured to execute the method comprising:

[0094] (S1) collecting a sequence of locust activity images in a target area by a multispectral camera device, and synchronously obtaining geographic coordinates and environmental parameters;

[0095] (S2) performing moving target detection on the image sequence, fusing background difference method and optical flow method to separate dynamic locust targets;

[0096] (S3) using the improved YOLOv8 deep learning model to identify and classify locust individuals, and output target location and category information;

[0097] (S4) Statistical modeling of locust populations in high-density overlapping areas based on density map regression networks;

[0098] (S5) Combined with geographic information system (GIS) to generate three-dimensional population density heat maps and predict migration paths;

[0099] The memory stores a locust feature database and model parameters.

[0100] The present invention also provides a computer-readable storage medium storing program instructions, which, when executed by a processor, comprises the following steps:

[0101] (S1) collecting a sequence of locust activity images in a target area by a multispectral camera device, and synchronously obtaining geographic coordinates and environmental parameters;

[0102] (S2) performing moving target detection on the image sequence, fusing background difference method and optical flow method to separate dynamic locust targets;

[0103] (S3) using the improved YOLOv8 deep learning model to identify and classify locust individuals, and output target location and category information;

[0104] (S4) Statistical modeling of locust populations in high-density overlapping areas based on density map regression networks;

[0105] (S5) Combined with geographic information system (GIS), three-dimensional population density heat maps were generated and migration paths were predicted.

[0106] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring method for locust population density based on AI visual recognition, characterized by: The following steps are involved: (S1) collecting a sequence of locust activity images in a target area by a multispectral camera device, and synchronously obtaining geographic coordinates and environmental parameters; (S2) performing moving target detection on the image sequence, fusing background difference method and optical flow method to separate dynamic locust targets; (S3) using the improved YOLOv8 deep learning model to identify and classify locust individuals, and output target location and category information; (S4) Statistical modeling of locust populations in high-density overlapping areas based on density map regression networks; (S5) Combined with geographic information system to generate three-dimensional population density heat map and predict migration path.

2. The method for intelligently monitoring locust population density based on AI visual recognition according to claim 1, characterized in that: In step S1: the multispectral camera device includes a visible light band, a near infrared band and a polarized light imaging module; Image acquisition is completed through the collaboration of drones and ground fixed monitoring stations.

3. The method for intelligent monitoring of locust population density based on AI visual recognition according to claim 1, characterized in that: The step S2 specifically includes: Build a dynamic background model based on the ViBe algorithm to separate the foreground moving targets; The pyramid Lucas-Kanade optical flow method is used to calculate the locust motion vector; The motion trajectory is aggregated and analyzed using the DBSCAN clustering algorithm to eliminate noise interference.

4. The method for intelligently monitoring locust population density based on AI visual recognition according to claim 1, characterized in that: The improved YOLOv8 model includes the following optimizations: The backbone network is replaced with the ConvNeXt-Tiny structure to improve the detection capability of small objects; The anchor frame parameters are optimized based on locust morphological data, and three scales are set: 8×8, 16×16, and 32×32; The detection head integrates the CBAM attention mechanism, and the weight calculation formula is: σ(W1·ReLU(W0·Favg)+W1·ReLU(W0·Fmax)) Among them: Favg and Fmax are average pooling and maximum pooling features respectively.

5. The method for intelligent monitoring of locust population density based on AI visual recognition according to claim 1, characterized in that: The density map regression network adopts: Lightweight CSRNet architecture, consisting of 10 dilated convolutional layers with a dilation rate of 2; Adaptive spatial transformation module, the formula is defined as: Used to correct perspective distortion; Multi-scale feature fusion branch, the number of output channels is 256.

6. The method for intelligently monitoring locust population density based on AI visual recognition according to claim 1, characterized in that: The GIS system implementation in step S5 includes: Generate density distribution surface based on Kriging interpolation method; Combining wind speed and direction data to build a migration and diffusion model, the differential equation is: Where: D is the diffusion coefficient, v is the migration velocity, and λ is the population growth rate; The disaster levels are divided into level one, level two, and level three according to the density threshold.

7. The method for intelligently monitoring locust population density based on AI visual recognition according to claim 1, characterized in that: Also includes: The edge computing node executes steps S2-S4 and uses TensorRT to accelerate inference; Transmit density data back to the cloud analysis platform via the 5G NR network; The monitoring records are encrypted with SHA-256 hash and written into the blockchain for evidence storage.

8. The method for intelligently monitoring locust population density based on AI visual recognition according to claim 1, wherein the locust classification comprises: Morphological distinction between adults and nymphs; The confidence threshold of suspected targets is set to 0.7; Mating behavior recognition is achieved through a posture estimation model.

9. An electronic device, characterized in that include: Multispectral image acquisition module; A processor configured to execute the method according to any one of claims 1 to 8; The memory stores a locust feature database and model parameters.

10. A computer-readable storage medium, characterized in that Program instructions are stored, and when executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images