A dual-spectral visual method and system for monitoring epidemic targets

Abstract

This invention belongs to the field of diseased tree monitoring technology. It discloses a dual-spectral visual method and system for monitoring diseased tree targets. The method includes the following steps: constructing a dual-spectral diseased tree target identification model, a three-dimensional forest model, and a diseased tree target path planning model; using the dual-spectral diseased tree target identification model to identify diseased trees; using the diseased tree target path planning model to plan the path for diseased trees; and adding the real-time total path of the diseased tree target to the three-dimensional forest model with a forest spatial rectangular coordinate system. The system includes a dual-spectral diseased tree target identification model unit, a real-time dual-spectral diseased tree image acquisition unit, a three-dimensional forest model unit, a forest spatial rectangular coordinate system establishment unit, a diseased tree target path display unit, and a diseased tree target path planning model unit. This invention solves the problems of high labor costs, heavy workload, low monitoring efficiency, low security, and high experience requirements for personnel in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of pest monitoring technology, specifically relating to a dual-spectrum visual pest target monitoring method and system. Background Technology

[0002] Once a forest is infected with an epidemic, without human intervention, it often spreads rapidly and over a large area due to high tree density and cross-infection, causing enormous losses to the economy and the ecological environment. For example, pine wilt disease caused by pine nematode is known as the cancer of pine trees; it is a devastating disease characterized by multiple transmission routes, hidden infection sites, rapid onset, long incubation period, and great difficulty in control. Under natural conditions, pine wilt disease can affect 45 species of pine trees. Once infected, pine trees can die in as little as 40 days, and without human intervention, it can destroy entire forests within 3 to 5 years. Therefore, the key to controlling pine wilt disease lies in monitoring and prevention. Only through timely detection and thorough eradication can the generation of new sources of infection be cut off and the spread of the disease suppressed.

[0003] Existing methods for monitoring and controlling diseased trees mostly rely on manual forest patrols, which involve high labor costs, a large workload, and low monitoring efficiency. The complex environment and rugged terrain of forests pose risks to the safety of personnel. Furthermore, forest patrols require personnel to be very familiar with forest paths in order to select the safest and fastest routes, thus demanding a high level of experience from the staff. Summary of the Invention

[0004] To address the problems of high labor costs, heavy workload, low monitoring efficiency, low security, and high requirements for staff experience in existing technologies, this invention aims to provide a dual-spectrum visual target monitoring method and system.

[0005] The technical solution adopted in this invention is as follows:

[0006] A dual-spectral visual method for detecting infectious targets includes the following steps:

[0007] A dataset of historical bispectral images of diseased trees in different forest scenes was obtained, and a bispectral image recognition model for diseased trees was constructed based on the dataset using image recognition algorithms.

[0008] Obtain 3D forest data of the target forest, and construct a 3D forest model of the target forest using 3D modeling algorithms based on the 3D forest data;

[0009] Based on the target forest's three-dimensional forest model, a forest spatial rectangular coordinate system is established, resulting in a forest three-dimensional model with the forest spatial rectangular coordinate system set.

[0010] Obtain the spatial coordinates of all accessible sub-paths in a 3D forest model with a forest spatial rectangular coordinate system.

[0011] Based on the spatial coordinates of the sub-paths of all passable paths, a swarm intelligence optimization algorithm is used to construct a target path planning model for the epidemic.

[0012] In a 3D forest model with a rectangular coordinate system, several monitoring points are added, and the spatial coordinates of each monitoring point are obtained.

[0013] Real-time bispectral image data of infected trees were collected from all monitoring points, and the infected tree target was identified using a bispectral infected tree target recognition model based on the real-time bispectral infected tree image data, thus obtaining the real-time infected tree target recognition results.

[0014] If any real-time identification result of an infected tree target indicates the presence of an infected tree target, then the corresponding monitoring point will be designated as the infected tree target point.

[0015] Based on the spatial coordinates of the monitoring points of all infected tree target points and the spatial coordinates of the sub-paths of all passable paths, the infected tree target path planning model is used to plan the infected tree target path and obtain the real-time total path of the infected tree target.

[0016] Add the real-time total path of the infected trees to the forest 3D model with a forest spatial rectangular coordinate system to obtain a forest 3D model that displays the real-time total path of the infected trees.

[0017] Furthermore, a historical bispectral image dataset of infected trees in different forest scenes is obtained, and based on this dataset, an image recognition algorithm is used to construct a bispectral infected tree target recognition model, including the following steps:

[0018] Image preprocessing was performed on each historical bispectral infected tree image in the historical bispectral infected tree image dataset to obtain the preprocessed historical bispectral infected tree image dataset.

[0019] The preprocessed historical bispectral infected tree image dataset was divided into a bispectral infected tree target recognition training sample set and a bispectral infected tree target recognition test sample set.

[0020] Based on the training sample set for dual-spectral diseased tree target recognition, the LFTCNN algorithm of image recognition algorithm is used for training and optimization to construct an initial dual-spectral diseased tree target recognition model;

[0021] Based on the dual-spectral target recognition test sample set, the initial dual-spectral target recognition model is trained and tested. If the test accuracy is greater than the threshold, the optimal dual-spectral target recognition model is output; otherwise, training and optimization continue.

[0022] Furthermore, the historical / real-time dual-spectral infected tree image data includes corresponding historical / real-time infrared infected tree image data and historical / real-time visible light infected tree image data.

[0023] Furthermore, the dual-spectral target recognition model for epidemic pests includes an input layer, a first convolutional channel, a second convolutional channel, a fully connected layer, a classification layer based on an Elman neural network, and an output layer. The input layer is connected to the first and second convolutional channels respectively, and both the first and second convolutional channels are connected to the fully connected layer. The fully connected layer, the classification layer, and the output layer are connected sequentially.

[0024] The first convolutional channel includes a plurality of first convolutional modules connected in sequence. Each first convolutional module includes a first convolutional layer and a first pooling layer connected in sequence. The first pooling layer at the beginning is also connected to a fully connected layer.

[0025] The second convolution channel includes several second convolution modules connected in sequence. Each second convolution module includes a second convolution layer and a second pooling layer connected in sequence. The second pooling layer at the beginning is also connected to a fully connected layer.

[0026] Furthermore, the 3D modeling algorithm is a polygon mesh 3D modeling algorithm;

[0027] The forest spatial rectangular coordinate system includes mutually perpendicular X-axis, Y-axis and Z-axis in space, and the spatial position coordinates of the forest spatial rectangular coordinate system are in the format of (x,y,z).

[0028] Furthermore, the passable path includes a sub-path start point and a sub-path end point, and the spatial coordinates of the sub-path of the passable path include the spatial coordinates of the sub-path start point and the corresponding spatial coordinates of the sub-path end point.

[0029] Furthermore, based on the spatial coordinates of the sub-paths of all passable paths, a swarm intelligence optimization algorithm is used to construct a target path planning model for the epidemic, including the following steps:

[0030] Based on a three-dimensional forest model with a forest spatial rectangular coordinate system, several historical total path starting point spatial coordinates and several historical total path ending point spatial coordinates are randomly set.

[0031] Randomly select the spatial coordinates of the starting point and the ending point of the historical total path. Based on these spatial coordinates, as well as the spatial coordinates of the starting point and the ending point of all passable paths, the initial historical epidemic target total path is obtained.

[0032] The spatial coordinates of the starting point of the initial historical total path of the target of the epidemic, the number of sub-paths, and the spatial coordinates of the starting point of each sub-path are used as the positions of individual whales in the IWOA population;

[0033] The IWOA optimization algorithm is used to surround, search and bubble net attack the IWOA population, update the IWOA population and obtain the updated IWOA population.

[0034] Output the position of the global optimal solution corresponding to the best whale individual in the updated IWOA population, and obtain the optimal historical total path starting point spatial coordinates, the optimal number of sub-paths, and the optimal sub-path starting point spatial coordinates of each sub-path.

[0035] The optimal historical total path for the target of the epidemic is obtained based on the optimal spatial coordinates of the starting point of the initial historical total path, the optimal number of sub-paths, and the optimal spatial coordinates of the starting point of each sub-path.

[0036] The spatial coordinates of the starting point and ending point of the historical total path are updated to obtain the updated spatial coordinates of the starting point and ending point of the historical total path. Based on the updated spatial coordinates of the starting point and ending point of the historical total path, as well as the spatial coordinates of the starting point and the corresponding ending point of all passable paths, the IWOA optimization algorithm is used for optimization training to construct the target path planning model for the epidemic.

[0037] Furthermore, real-time bispectral image data of infected trees from all monitoring points were collected, and based on this data, a bispectral infected tree target recognition model was used to identify infected tree targets, yielding real-time infected tree target recognition results. This process included the following steps:

[0038] Real-time dual-spectral image data of infected trees were collected from all monitoring points; the real-time dual-spectral image data of infected trees included real-time infrared image data and real-time visible light image data of infected trees.

[0039] Using the first convolutional channel of the dual-spectral infected wood target recognition model, low-level features and high-level features of real-time dual-spectral infected wood image data are extracted from real-time infrared infected wood image data.

[0040] The second convolutional channel of the dual-spectral infected wood target recognition model is used to extract low-level features and high-level features of real-time visible light infected wood image data from real-time dual-spectral infected wood image data.

[0041] Feature fusion was performed on the low-level features, high-level features, low-level features, and high-level features of real-time infrared infected wood image data, to obtain the corresponding real-time dual-spectral infected wood image data fusion features.

[0042] Based on the fusion features of real-time dual-spectral infected tree image data, classification is performed, and corresponding real-time infected tree target identification labels are output;

[0043] Based on the real-time epidemic tree target identification labels, the corresponding real-time epidemic tree target identification results are output.

[0044] Furthermore, the Whale Optimization Algorithm is improved by introducing Circle chaotic sequence initialization, dynamic reverse learning strategy and convergence factor, resulting in the IWOA Optimization Algorithm;

[0045] Based on the spatial coordinates of all monitoring points and sub-paths of all passable routes for the target trees, the target tree path planning model is used to plan the real-time total path for the target trees, including the following steps:

[0046] Randomly select the starting spatial coordinates of the real-time total path, and based on the monitoring point spatial coordinates from the starting spatial coordinates of the real-time total path to the last infected tree target point, as well as the starting spatial coordinates of all passable sub-paths and the corresponding ending spatial coordinates of the sub-paths, obtain the initial real-time infected tree target total path.

[0047] The initial real-time total path of the target target, the spatial coordinates of the starting point, the number of sub-paths, and the spatial coordinates of the starting point of each sub-path are used as the positions of individual whales in the IWOA population.

[0048] Initialize the IWOA optimization algorithm parameters and initialize the IWOA population using the Circle chaotic sequence;

[0049] Calculate the fitness value of each whale individual in the IWOA population, and retain the best whale individual based on the fitness value of each whale individual;

[0050] A random parameter p is generated. If p < 0.5 and |A| < 1, the prey-encircling behavior improved by the convergence factor is executed, and the IWOA population position is updated. If p < 0.5 and |A| ≥ 1, the prey-searching behavior improved by the convergence factor is executed, and the IWOA population position is updated. If p ≥ 0.5, the bubble-net attack behavior improved by the convergence factor is executed, and the IWOA population position is updated. Here, p is the update parameter, and A is the step size coefficient of the convergence factor optimization.

[0051] Based on the updated IWOA population, dynamic reverse learning is performed to obtain a reverse IWOA population. The fitness value of each whale individual in the updated IWOA population and the reverse IWOA population is calculated. Based on the fitness values ​​of all whale individuals, the optimal whale individual is updated to obtain the updated optimal whale individual.

[0052] Determine whether the number of iterations meets the requirement or whether the optimal fitness value corresponding to the updated optimal whale individual meets the requirement. If yes, output the position of the global optimal solution corresponding to the updated optimal whale individual, and obtain the optimal real-time total path starting spatial coordinates, the optimal number of sub-paths, and the optimal sub-path starting spatial coordinates of each sub-path. Otherwise, perform the next update of the IWOA population.

[0053] The optimal real-time total path for the epidemic target is obtained based on the optimal spatial coordinates of the starting point of the initial real-time total path, the optimal number of sub-paths, and the optimal spatial coordinates of the starting point of each sub-path.

[0054] A dual-spectral visual diseased tree target monitoring system is provided to implement a dual-spectral visual diseased tree target monitoring method. The system includes a dual-spectral diseased tree target recognition model construction unit, a forest 3D model construction unit, a forest spatial rectangular coordinate system establishment unit, a diseased tree target path planning model construction unit, a real-time dual-spectral diseased tree image acquisition unit, a dual-spectral diseased tree target recognition model application unit, a diseased tree target path planning model application unit, and a diseased tree target path display unit. The dual-spectral diseased tree target recognition model construction unit, the forest 3D model construction unit, the forest spatial rectangular coordinate system establishment unit, the diseased tree target path planning model construction unit, the real-time dual-spectral diseased tree image acquisition unit, the dual-spectral diseased tree target recognition model application unit, the diseased tree target path planning model application unit, and the diseased tree target path display unit are connected sequentially.

[0055] The dual-spectral diseased tree target recognition model building unit is used to acquire historical dual-spectral diseased tree image datasets in different forest scenes, and to construct a dual-spectral diseased tree target recognition model based on the historical dual-spectral diseased tree image datasets using image recognition algorithms.

[0056] The forest 3D model building unit is used to acquire the 3D data of the target forest and, based on the 3D data, to construct the 3D model of the target forest using 3D modeling algorithms.

[0057] The forest spatial rectangular coordinate system establishment unit is used to establish a forest spatial rectangular coordinate system based on a target forest 3D model, thereby obtaining a forest 3D model with the forest spatial rectangular coordinate system set; to obtain the spatial location coordinates of all accessible sub-paths in the forest 3D model with the forest spatial rectangular coordinate system set; and to add several monitoring points in the forest 3D model with the forest spatial rectangular coordinate system set, and obtain the monitoring point spatial location coordinates of each monitoring point.

[0058] The target path planning model construction unit for infected trees is used to construct the target path planning model for infected trees based on the spatial coordinates of the sub-paths of all passable paths and using a swarm intelligence optimization algorithm.

[0059] The real-time dual-spectral diseased tree image acquisition unit is set at the actual location of the corresponding monitoring point in the forest to collect real-time dual-spectral diseased tree image data from all monitoring points.

[0060] The dual-spectral diseased tree target recognition model application unit is used to identify diseased trees based on real-time dual-spectral diseased tree image data using the dual-spectral diseased tree target recognition model, and obtain real-time diseased tree target recognition results; it determines whether any real-time diseased tree target recognition result indicates the existence of a diseased tree target, and if any real-time diseased tree target recognition result indicates the existence of a diseased tree target, then the corresponding monitoring point is taken as the diseased tree target point;

[0061] The application unit of the infected tree target path planning model is used to plan the infected tree target path based on the spatial coordinates of the monitoring points of all infected tree target points and the spatial coordinates of the sub-paths of all passable paths, and obtain the real-time total path of the infected tree target.

[0062] The Diseased Tree Target Path Display Unit is used to add the real-time total path of the diseased tree target to a forest 3D model with a forest spatial rectangular coordinate system, thereby obtaining a forest 3D model that displays the real-time total path of the diseased tree target.

[0063] The beneficial effects of this invention are as follows:

[0064] This invention provides a dual-spectrum visual method and system for monitoring infected trees. By constructing a dual-spectrum infected tree target recognition model, it can achieve online, real-time, and accurate identification of infected tree targets, avoiding manual inspection, reducing labor costs, decreasing workload, and improving monitoring efficiency. By constructing a three-dimensional forest model of the target forest, it improves the staff's understanding of the target forest and provides a reference for outdoor operations. Finally, by constructing an infected tree target path planning model to obtain the real-time total path of infected tree targets and adding the real-time total path of infected tree targets to the three-dimensional forest model with a forest spatial rectangular coordinate system, it reduces the experience requirements of the staff and improves the safety of outdoor work.

[0065] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0066] Figure 1 This is a flowchart of the dual-spectral visual target monitoring method in this invention.

[0067] Figure 2 This is a structural block diagram of the dual-spectrum visual target monitoring system of the present invention. Detailed Implementation

[0068] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0069] Example 1:

[0070] like Figure 1 As shown, this embodiment provides a dual-spectral visual target monitoring method, including the following steps:

[0071] A historical dual-spectral image dataset of infected trees in different forest scenes was obtained. Based on this dataset, an image recognition algorithm was used to construct a dual-spectral image dataset for identifying infected trees. The process included the following steps:

[0072] Image preprocessing is performed on each historical dual-spectral infected tree image data in the historical dual-spectral infected tree image dataset to obtain the preprocessed historical dual-spectral infected tree image dataset; the historical dual-spectral infected tree image data includes the corresponding historical infrared infected tree image data and historical visible light infected tree image data.

[0073] The preprocessed historical bispectral infected tree image dataset was divided into a bispectral infected tree target recognition training sample set and a bispectral infected tree target recognition test sample set.

[0074] Based on the training sample set for dual-spectral epidemic wood target recognition, the low-level feature fusion two-channel convolutional neural network (LFTCNN) algorithm of image recognition algorithm is used for training and optimization to construct the initial dual-spectral epidemic wood target recognition model.

[0075] By using dual convolutional channels to extract features separately from historical dual-spectral infected tree image data, including corresponding historical infrared infected tree image data and historical visible light infected tree image data, and then performing feature fusion, the accuracy of infected tree target identification was improved.

[0076] The dual-spectral target recognition model for infected trees includes an input layer, a first convolutional channel, a second convolutional channel, a fully connected layer, a classification layer based on an Elman neural network, and an output layer. The input layer is connected to the first and second convolutional channels, and both the first and second convolutional channels are connected to the fully connected layer. The fully connected layer, the classification layer, and the output layer are connected sequentially.

[0077] The first convolutional channel includes a plurality of first convolutional modules connected in sequence. Each first convolutional module includes a first convolutional layer and a first pooling layer connected in sequence. The first pooling layer at the beginning is also connected to a fully connected layer.

[0078] The second convolution channel includes a number of second convolution modules connected in sequence. Each second convolution module includes a second convolution layer and a second pooling layer connected in sequence. The second pooling layer at the beginning is also connected to a fully connected layer.

[0079] The input layer is used to receive historical / real-time bispectral infected tree image data, transmit historical / real-time infrared infected tree image data from the historical / real-time bispectral infected tree image data to the first convolutional channel, and transmit historical / real-time visible light infected tree image data to the second convolutional channel;

[0080] The first convolutional channel is used to receive historical / real-time infrared diseased tree image data transmitted from the input layer, extract features from the historical / real-time infrared diseased tree image data to obtain low-level features and high-level features of the historical / real-time infrared diseased tree image data, and transmit the low-level features and high-level features of the historical / real-time infrared diseased tree image data to the fully connected layer.

[0081] The first convolutional module is used to extract features from historical / real-time infrared diseased tree image data. The first convolutional module at the beginning extracts low-level features from the historical / real-time infrared diseased tree image data and transmits the low-level features to the fully connected layer. The first convolutional module at the end extracts high-level features from the historical / real-time infrared diseased tree image data and transmits the high-level features to the fully connected layer.

[0082] The first convolutional layer is used to convolve the historical / real-time infrared epidemic tree image data or the low-level features of the historical / real-time infrared epidemic tree image data.

[0083] The first pooling layer is used to pool the low-level features of historical / real-time infrared epidemic image data;

[0084] The second convolutional channel is used to receive historical / real-time visible light infected tree image data transmitted from the input layer, extract features from the historical / real-time visible light infected tree image data to obtain low-level features and high-level features of the historical / real-time visible light infected tree image data, and transmit the low-level features and high-level features of the historical / real-time visible light infected tree image data to the fully connected layer.

[0085] The second convolutional module is used to extract features from historical / real-time visible light infected tree image data. The first convolutional module extracts low-level features from the historical / real-time visible light infected tree image data and transmits these low-level features to the fully connected layer. The last convolutional module extracts high-level features from the historical / real-time visible light infected tree image data and transmits these high-level features to the fully connected layer.

[0086] The second convolutional layer is used to convolve historical / real-time visible light infected tree image data or low-level features of historical / real-time visible light infected tree image data;

[0087] The second pooling layer is used to pool the low-level features of historical / real-time visible light epidemic image data;

[0088] A fully connected layer is used to receive low-level and high-level features of historical / real-time infrared diseased tree image data transmitted through the first convolutional channel, as well as low-level and high-level features of historical / real-time visible light diseased tree image data transmitted through the second convolutional channel. It performs feature fusion on the low-level, high-level, and low-level features of historical / real-time infrared diseased tree image data, as well as the high-level features of historical / real-time visible light diseased tree image data, to obtain the corresponding historical / real-time dual-spectral diseased tree image data fused features, and then transmits the historical / real-time dual-spectral diseased tree image data fused features to the classification layer.

[0089] The classification layer receives the fusion features of historical / real-time bispectral infected tree image data transmitted from the fully connected layer. Based on the fusion features of historical / real-time bispectral infected tree image data, it performs classification using a pre-trained Elman neural network structure, outputs the corresponding historical / real-time infected tree target identification labels, and transmits the historical / real-time infected tree target identification labels to the output layer.

[0090] The output layer receives historical / real-time epidemic target identification tags transmitted from the classification layer and outputs the corresponding historical / real-time epidemic target identification results based on the historical / real-time epidemic target identification tags.

[0091] Based on the dual-spectral epidemic tree target recognition test sample set, the initial dual-spectral epidemic tree target recognition model is trained and tested. If the test accuracy is greater than the threshold, the optimal dual-spectral epidemic tree target recognition model is output; otherwise, training and optimization continue.

[0092] Obtain 3D forest data of the target forest, and construct a 3D forest model of the target forest using a 3D modeling algorithm based on the 3D forest data; the 3D modeling algorithm is a polygon mesh 3D modeling algorithm.

[0093] Based on the target forest's 3D model, a forest spatial rectangular coordinate system is established, resulting in a 3D forest model with this system. The forest spatial rectangular coordinate system includes mutually perpendicular X, Y, and Z axes in space, and the spatial coordinates of the forest spatial rectangular coordinate system are in the format (x, y, z). This system can transform path planning from the original two-dimensional space to a three-dimensional space, making the obtained real-time target path more consistent with the actual forest conditions and improving the accuracy of path planning.

[0094] This method obtains the spatial coordinates of all accessible sub-paths in a 3D forest model with a forest spatial rectangular coordinate system. Each accessible path includes a sub-path start point and a sub-path end point, and the spatial coordinates of the accessible sub-paths include the spatial coordinates of the sub-path start point and the corresponding sub-path end point. This method avoids duplication and detours of accessible paths and improves the accuracy of path planning.

[0095] Based on the spatial coordinates of the sub-paths of all passable paths, a swarm intelligence optimization algorithm is used to construct a target path planning model for the epidemic, including the following steps:

[0096] Based on a three-dimensional forest model with a forest spatial rectangular coordinate system, several historical total path starting point spatial coordinates and several historical total path ending point spatial coordinates are randomly set.

[0097] Randomly select the spatial coordinates of the starting point and the ending point of the historical total path. Based on these spatial coordinates, as well as the spatial coordinates of the starting point and the ending point of all passable paths, the initial historical epidemic target total path is obtained.

[0098] The spatial coordinates of the starting point of the initial historical total path of the target of the epidemic, the number of sub-paths, and the spatial coordinates of the starting point of each sub-path are used as the positions of individual whales in the IWOA population;

[0099] The Whale Optimization Algorithm (IWOA) is improved by introducing Circle chaotic sequence initialization, dynamic back learning strategy and convergence factor.

[0100] The IWOA optimization algorithm is used to surround, search and bubble net attack the IWOA population, update the IWOA population and obtain the updated IWOA population.

[0101] Output the position of the global optimal solution corresponding to the best whale individual in the updated IWOA population, and obtain the optimal historical total path starting point spatial coordinates, the optimal number of sub-paths, and the optimal sub-path starting point spatial coordinates of each sub-path.

[0102] The optimal historical total path for the target of the epidemic is obtained based on the optimal spatial coordinates of the starting point of the initial historical total path, the optimal number of sub-paths, and the optimal spatial coordinates of the starting point of each sub-path.

[0103] Update the spatial coordinates of the starting point and the ending point of the historical total path to obtain the updated spatial coordinates of the starting point and the ending point of the historical total path.

[0104] Based on the updated historical total path start point spatial coordinates and the updated historical total path end point spatial coordinates, as well as the start point spatial coordinates and corresponding end point spatial coordinates of all passable paths' sub-paths, the IWOA optimization algorithm is used for optimization training to construct the epidemic target path planning model.

[0105] In a 3D forest model with a rectangular coordinate system, several monitoring points are added, and the spatial coordinates of each monitoring point are obtained.

[0106] Real-time bispectral image data of infected trees were collected from all monitoring points. Based on the real-time bispectral image data, the bispectral infected tree target recognition model was used to identify infected tree targets and obtain real-time infected tree target recognition results. The process includes the following steps:

[0107] Real-time dual-spectral image data of infected trees were collected from all monitoring points; the real-time dual-spectral image data of infected trees included real-time infrared image data and real-time visible light image data of infected trees.

[0108] Using the first convolutional channel of the dual-spectral infected wood target recognition model, low-level features and high-level features of real-time dual-spectral infected wood image data are extracted from real-time infrared infected wood image data.

[0109] The second convolutional channel of the dual-spectral infected wood target recognition model is used to extract low-level features and high-level features of real-time visible light infected wood image data from real-time dual-spectral infected wood image data.

[0110] Feature fusion was performed on the low-level features, high-level features, low-level features, and high-level features of real-time infrared infected wood image data, to obtain the corresponding real-time dual-spectral infected wood image data fusion features.

[0111] Based on the fusion features of real-time dual-spectral infected tree image data, classification is performed, and corresponding real-time infected tree target identification labels are output;

[0112] Based on the real-time diseased tree target identification labels, output the corresponding real-time diseased tree target identification results;

[0113] If any real-time identification result of an infected tree target indicates the presence of an infected tree target, then the corresponding monitoring point will be designated as the infected tree target point.

[0114] Based on the spatial coordinates of all monitoring points and sub-paths of all passable routes for the target trees, the target tree path planning model is used to plan the real-time total path for the target trees, including the following steps:

[0115] Randomly select the starting spatial coordinates of the real-time total path, and based on the monitoring point spatial coordinates from the starting spatial coordinates of the real-time total path to the last infected tree target point, as well as the starting spatial coordinates of all passable sub-paths and the corresponding ending spatial coordinates of the sub-paths, obtain the initial real-time infected tree target total path.

[0116] The initial real-time total path of the target target, the spatial coordinates of the starting point, the number of sub-paths, and the spatial coordinates of the starting point of each sub-path are used as the positions of individual whales in the IWOA population.

[0117] Initialize the IWOA optimization algorithm parameters and initialize the IWOA population using the Circle chaotic sequence;

[0118] The formula for initializing a Circle chaotic sequence is:

[0119]

[0120] In the formula, x i+1,j+1 The initial positions of the whale population generated by the Circle chaotic mapping; x i,jis the initial position of the randomly generated whale population; mod(·) is the mod function; i is the individual whale indicator; j is the dimension indicator; compared with the randomly distributed population, the initial position of the improved population generated by mapping is more evenly distributed, which expands the search range of the whale population in space, increases the diversity of the population position, and improves the defect of the algorithm being prone to getting trapped in local extrema to a certain extent, thereby improving the optimization efficiency of the algorithm.

[0121] Calculate the fitness value of each whale individual in the IWOA population, and retain the best whale individual based on the fitness value of each whale individual;

[0122] The formula for fitness value is:

[0123]

[0124] In the formula, fit is the fitness function; E is the network output error function; y n This represents the actual output value of the nth network node. Let n be the ideal output value of the nth network node; n is the network node indicator; N is the total number of network nodes.

[0125] A random parameter p is generated. If p < 0.5 and |A| < 1, the prey-encircling behavior improved by the convergence factor is executed, and the IWOA population position is updated. If p < 0.5 and |A| ≥ 1, the prey-searching behavior improved by the convergence factor is executed, and the IWOA population position is updated. If p ≥ 0.5, the bubble-net attack behavior improved by the convergence factor is executed, and the IWOA population position is updated. Here, p is the update parameter, and A is the step size coefficient of the convergence factor optimization.

[0126] The formula for the prey-surrounding behavior is:

[0127] X 1 (t+1)=X * (t)-AD

[0128] In the formula, X 1 (t+1) represents the individual whale position updated by the prey-surrounding behavior; X * (t) represents the optimal position of the whale individual; D is the distance between the current whale individual and the optimal whale individual; A = 2ar - a, where a is the convergence factor decreasing from 2 to 0;

[0129] The formula for the convergence factor is:

[0130]

[0131] In the formula, a is the convergence factor; tanh(.) is the hyperbolic tangent function; t, t max These are the current iteration count and the maximum iteration count, respectively; amax a min λ and k are the maximum and minimum values ​​of the convergence factor, respectively; λ is the deceleration rate parameter, k is the deceleration period parameter, λ = -2π, k = π;

[0132] In the early stages of iteration, the value of a is relatively large, and the updated A is also relatively large, |A|≥1, which causes the IWOA algorithm to be in the prey-searching behavior for a long time in the early stages of iteration, enhancing the algorithm's global search capability. In the early stages of iteration, the value of a is relatively small, and the updated A is also relatively small, |A|<1, which causes the IWOA algorithm to be in the prey-encircling behavior for a long time in the early stages of iteration, enhancing the algorithm's local encirclement capability and improving its local hunting capability.

[0133] The formula for hunting behavior is:

[0134] X 2 (t+1)=X rand (t)-AD

[0135] In the formula, X 2 (t+1) represents the updated location of the individual whale based on its prey-hunting behavior; X rand (t) represents the location of a random whale individual selected from the IWOA population;

[0136] The formula for the attack behavior of Popo.net is:

[0137] X 3 (t+1)=D'c bl cos(2πl)+X * (t)

[0138] In the formula, X 3 (t+1) represents the position of the whale individual updated by the bubble net attack behavior; D' represents the distance between the current whale individual and the prey; b is a constant defining the spiral equation, b = 1; l is a random number between [-1, 1];

[0139] Based on the updated IWOA population, dynamic reverse learning is performed to obtain a reverse IWOA population. The fitness value of each whale individual in the updated IWOA population and the reverse IWOA population is calculated. Based on the fitness values ​​of all whale individuals, the optimal whale individual is updated to obtain the updated optimal whale individual.

[0140] The formula for the dynamic back-learning strategy is:

[0141] X' ij (t)=k(a j (t)+b j (t))-X ij (t)

[0142] In the formula, X' ij (t), Xij (t) represents the reverse solution position and the forward solution position of the i-th individual in the j-th dimension, respectively; a j (t), b j (t) represents the upper and lower bounds of the j-th dimension of the current IWOA population, respectively; k is the decreasing inertia factor, k = 0.9 - 0.5t / t max ;t、t max These represent the current iteration count and the maximum iteration count, respectively; this reduces search blind spots and more effectively avoids premature convergence and getting trapped in local optima.

[0143] Determine whether the number of iterations meets the requirement or whether the optimal fitness value corresponding to the updated optimal whale individual meets the requirement. If yes, output the position of the global optimal solution corresponding to the updated optimal whale individual, and obtain the optimal real-time total path starting spatial coordinates, the optimal number of sub-paths, and the optimal sub-path starting spatial coordinates of each sub-path. Otherwise, perform the next update of the IWOA population.

[0144] The optimal real-time total path of the epidemic target is obtained based on the optimal spatial coordinates of the starting point of the initial real-time total path, the optimal number of sub-paths, and the optimal spatial coordinates of the starting point of each sub-path.

[0145] Add the real-time total path of the infected trees to the forest 3D model with a forest spatial rectangular coordinate system to obtain a forest 3D model that displays the real-time total path of the infected trees.

[0146] Example 2:

[0147] like Figure 2 As shown, this embodiment provides a dual-spectrum visual diseased tree target monitoring system for implementing a dual-spectrum visual diseased tree target monitoring method. The system includes a dual-spectrum diseased tree target recognition model construction unit, a forest 3D model construction unit, a forest spatial rectangular coordinate system establishment unit, a diseased tree target path planning model construction unit, a real-time dual-spectrum diseased tree image acquisition unit, a dual-spectrum diseased tree target recognition model application unit, a diseased tree target path planning model application unit, and a diseased tree target path display unit. The dual-spectrum diseased tree target recognition model construction unit, the forest 3D model construction unit, the forest spatial rectangular coordinate system establishment unit, the diseased tree target path planning model construction unit, the real-time dual-spectrum diseased tree image acquisition unit, the dual-spectrum diseased tree target recognition model application unit, the diseased tree target path planning model application unit, and the diseased tree target path display unit are connected sequentially.

[0148] The dual-spectral diseased tree target recognition model building unit is used to acquire historical dual-spectral diseased tree image datasets in different forest scenes, and to construct a dual-spectral diseased tree target recognition model based on the historical dual-spectral diseased tree image datasets using image recognition algorithms.

[0149] The forest 3D model building unit is used to acquire the 3D data of the target forest and, based on the 3D data, to construct the 3D model of the target forest using 3D modeling algorithms.

[0150] The forest spatial rectangular coordinate system establishment unit is used to establish a forest spatial rectangular coordinate system based on a target forest 3D model, thereby obtaining a forest 3D model with the forest spatial rectangular coordinate system set; to obtain the spatial location coordinates of all accessible sub-paths in the forest 3D model with the forest spatial rectangular coordinate system set; and to add several monitoring points in the forest 3D model with the forest spatial rectangular coordinate system set, and obtain the monitoring point spatial location coordinates of each monitoring point.

[0151] The target path planning model construction unit for infected trees is used to construct the target path planning model for infected trees based on the spatial coordinates of the sub-paths of all passable paths and using a swarm intelligence optimization algorithm.

[0152] The real-time dual-spectral diseased tree image acquisition unit is set at the actual location of the corresponding monitoring point in the forest to collect real-time dual-spectral diseased tree image data from all monitoring points.

[0153] The dual-spectral diseased tree target recognition model application unit is used to identify diseased trees based on real-time dual-spectral diseased tree image data using the dual-spectral diseased tree target recognition model, and obtain real-time diseased tree target recognition results; it determines whether any real-time diseased tree target recognition result indicates the existence of a diseased tree target, and if any real-time diseased tree target recognition result indicates the existence of a diseased tree target, then the corresponding monitoring point is taken as the diseased tree target point;

[0154] The application unit of the infected tree target path planning model is used to plan the infected tree target path based on the spatial coordinates of the monitoring points of all infected tree target points and the spatial coordinates of the sub-paths of all passable paths, and obtain the real-time total path of the infected tree target.

[0155] The Diseased Tree Target Path Display Unit is used to add the real-time total path of the diseased tree target to a forest 3D model with a forest spatial rectangular coordinate system, thereby obtaining a forest 3D model that displays the real-time total path of the diseased tree target.

[0156] This invention provides a dual-spectrum visual method and system for monitoring infected trees. By constructing a dual-spectrum infected tree target recognition model, it can achieve online, real-time, and accurate identification of infected tree targets, avoiding manual inspection, reducing labor costs, decreasing workload, and improving monitoring efficiency. By constructing a three-dimensional forest model of the target forest, it improves the staff's understanding of the target forest and provides a reference for outdoor operations. Finally, by constructing an infected tree target path planning model to obtain the real-time total path of infected tree targets and adding the real-time total path of infected tree targets to the three-dimensional forest model with a forest spatial rectangular coordinate system, it reduces the experience requirements of the staff and improves the safety of outdoor work.

[0157] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A dual-spectral visual method for detecting infectious targets, characterized in that: Includes the following steps: A dataset of historical bispectral images of diseased trees in different forest scenes was obtained, and a bispectral image recognition model for diseased trees was constructed based on the dataset using image recognition algorithms. Obtain 3D forest data of the target forest, and construct a 3D forest model of the target forest using 3D modeling algorithms based on the 3D forest data; Based on the target forest's three-dimensional forest model, a forest spatial rectangular coordinate system is established, resulting in a forest three-dimensional model with the forest spatial rectangular coordinate system set. Obtain the spatial coordinates of all accessible sub-paths in a 3D forest model with a forest spatial rectangular coordinate system. Based on the spatial coordinates of the sub-paths of all passable paths, a swarm intelligence optimization algorithm is used to construct a target path planning model for the epidemic. In a 3D forest model with a rectangular coordinate system, several monitoring points are added, and the spatial coordinates of each monitoring point are obtained. Real-time bispectral image data of infected trees were collected from all monitoring points, and the infected tree target was identified using a bispectral infected tree target recognition model based on the real-time bispectral infected tree image data, thus obtaining the real-time infected tree target recognition results. If any real-time identification result of an infected tree target indicates the presence of an infected tree target, then the corresponding monitoring point will be designated as the infected tree target point. Based on the spatial coordinates of the monitoring points of all infected tree target points and the spatial coordinates of the sub-paths of all passable paths, the infected tree target path planning model is used to plan the infected tree target path and obtain the real-time total path of the infected tree target. Add the real-time total path of the infected trees to the forest 3D model with a forest spatial rectangular coordinate system to obtain a forest 3D model that displays the real-time total path of the infected trees.

2. The dual-spectral visual target monitoring method according to claim 1, characterized in that: A historical dual-spectral image dataset of infected trees in different forest scenes was obtained. Based on this dataset, an image recognition algorithm was used to construct a dual-spectral image dataset for identifying infected trees. The process included the following steps: Image preprocessing was performed on each historical bispectral infected tree image in the historical bispectral infected tree image dataset to obtain the preprocessed historical bispectral infected tree image dataset. The preprocessed historical bispectral infected tree image dataset was divided into a bispectral infected tree target recognition training sample set and a bispectral infected tree target recognition test sample set. Based on the training sample set for dual-spectral diseased tree target recognition, the LFTCNN algorithm of image recognition algorithm is used for training and optimization to construct an initial dual-spectral diseased tree target recognition model; Based on the dual-spectral target recognition test sample set, the initial dual-spectral target recognition model is trained and tested. If the test accuracy is greater than the threshold, the optimal dual-spectral target recognition model is output; otherwise, training and optimization continue.

3. The dual-spectral visual target monitoring method according to claim 2, characterized in that: The historical / real-time dual-spectral infected tree image data includes corresponding historical / real-time infrared infected tree image data and historical / real-time visible light infected tree image data.

4. The dual-spectral visual target monitoring method according to claim 3, characterized in that: The dual-spectral target recognition model for infected trees includes an input layer, a first convolutional channel, a second convolutional channel, a fully connected layer, a classification layer based on an Elman neural network, and an output layer. The input layer is connected to the first and second convolutional channels, and both the first and second convolutional channels are connected to the fully connected layer. The fully connected layer, the classification layer, and the output layer are connected sequentially. The first convolution channel includes a plurality of first convolution modules connected in sequence. Each first convolution module includes a first convolution layer and a first pooling layer connected in sequence. The first pooling layer at the beginning is also connected to a fully connected layer. The second convolution channel includes a plurality of second convolution modules connected in sequence. Each second convolution module includes a second convolution layer and a second pooling layer connected in sequence. The second pooling layer at the beginning is also connected to a fully connected layer.

5. The dual-spectral visual target monitoring method according to claim 1, characterized in that: The aforementioned 3D modeling algorithm is a polygon mesh 3D modeling algorithm; The aforementioned forest spatial rectangular coordinate system includes mutually perpendicular X-axis, Y-axis and Z-axis in space, and the spatial position coordinates of the forest spatial rectangular coordinate system are in the format of (x,y,z).

6. The dual-spectral visual target monitoring method according to claim 1, characterized in that: The passable path includes a sub-path start point and a sub-path end point, and the spatial coordinates of the sub-path of the passable path include the spatial coordinates of the sub-path start point and the corresponding spatial coordinates of the sub-path end point.

7. The dual-spectral visual target monitoring method according to claim 6, characterized in that: Based on the spatial coordinates of the sub-paths of all passable paths, a swarm intelligence optimization algorithm is used to construct a target path planning model for the epidemic, including the following steps: Based on a three-dimensional forest model with a forest spatial rectangular coordinate system, several historical total path starting point spatial coordinates and several historical total path ending point spatial coordinates are randomly set. Randomly select the spatial coordinates of the starting point and the ending point of the historical total path. Based on these spatial coordinates, as well as the spatial coordinates of the starting point and the ending point of all passable paths, the initial historical epidemic target total path is obtained. The spatial coordinates of the starting point of the initial historical total path of the target of the epidemic, the number of sub-paths, and the spatial coordinates of the starting point of each sub-path are used as the positions of individual whales in the IWOA population; The IWOA optimization algorithm is used to surround, search and bubble net attack the IWOA population, update the IWOA population and obtain the updated IWOA population. Output the position of the global optimal solution corresponding to the best whale individual in the updated IWOA population, and obtain the optimal historical total path starting point spatial coordinates, the optimal number of sub-paths, and the optimal sub-path starting point spatial coordinates of each sub-path. The optimal historical total path for the target of the epidemic is obtained based on the optimal spatial coordinates of the starting point of the initial historical total path, the optimal number of sub-paths, and the optimal spatial coordinates of the starting point of each sub-path. The spatial coordinates of the starting point and ending point of the historical total path are updated to obtain the updated spatial coordinates of the starting point and ending point of the historical total path. Based on the updated spatial coordinates of the starting point and ending point of the historical total path, as well as the spatial coordinates of the starting point and the corresponding ending point of all passable paths, the IWOA optimization algorithm is used for optimization training to construct the target path planning model for the epidemic.

8. The dual-spectral visual target monitoring method according to claim 4, characterized in that: Real-time bispectral image data of infected trees were collected from all monitoring points. Based on the real-time bispectral image data, the bispectral infected tree target recognition model was used to identify infected tree targets and obtain real-time infected tree target recognition results. The process includes the following steps: Real-time bispectral image data of infected trees were collected from all monitoring points; the real-time bispectral image data of infected trees included real-time infrared image data of infected trees and real-time visible light image data of infected trees. Using the first convolutional channel of the dual-spectral infected wood target recognition model, low-level features and high-level features of real-time dual-spectral infected wood image data are extracted from real-time infrared infected wood image data. The second convolutional channel of the dual-spectral infected wood target recognition model is used to extract low-level features and high-level features of real-time visible light infected wood image data from real-time dual-spectral infected wood image data. Feature fusion was performed on the low-level features, high-level features, low-level features, and high-level features of real-time infrared infected wood image data, to obtain the corresponding real-time dual-spectral infected wood image data fusion features. Based on the fusion features of real-time dual-spectral infected tree image data, classification is performed, and corresponding real-time infected tree target identification labels are output; Based on the real-time epidemic tree target identification labels, the corresponding real-time epidemic tree target identification results are output.

9. The dual-spectral visual target monitoring method according to claim 7, characterized in that: The Whale Optimization Algorithm is improved by introducing Circle chaotic sequence initialization, dynamic back learning strategy and convergence factor, resulting in the IWOA Optimization Algorithm; Based on the spatial coordinates of all monitoring points and sub-paths of all passable routes for the target trees, the target tree path planning model is used to plan the real-time total path for the target trees, including the following steps: Randomly select the starting spatial coordinates of the real-time total path, and based on the monitoring point spatial coordinates from the starting spatial coordinates of the real-time total path to the last infected tree target point, as well as the starting spatial coordinates of all passable sub-paths and the corresponding ending spatial coordinates of the sub-paths, obtain the initial real-time infected tree target total path. The initial real-time total path of the target target, the spatial coordinates of the starting point, the number of sub-paths, and the spatial coordinates of the starting point of each sub-path are used as the positions of individual whales in the IWOA population. Initialize the IWOA optimization algorithm parameters and initialize the IWOA population using the Circle chaotic sequence; Calculate the fitness value of each whale individual in the IWOA population, and retain the best whale individual based on the fitness value of each whale individual; A random parameter p is generated. If p < 0.5 and |A| < 1, the prey-encircling behavior improved by the convergence factor is executed, and the IWOA population position is updated. If p < 0.5 and |A| ≥ 1, the prey-searching behavior improved by the convergence factor is executed, and the IWOA population position is updated. If p ≥ 0.5, the bubble-net attack behavior improved by the convergence factor is executed, and the IWOA population position is updated. Here, p is the update parameter, and A is the step size coefficient of the convergence factor optimization. Based on the updated IWOA population, dynamic reverse learning is performed to obtain a reverse IWOA population. The fitness value of each whale individual in the updated IWOA population and the reverse IWOA population is calculated. Based on the fitness values ​​of all whale individuals, the optimal whale individual is updated to obtain the updated optimal whale individual. Determine whether the number of iterations meets the requirement or whether the optimal fitness value corresponding to the updated optimal whale individual meets the requirement. If yes, output the position of the global optimal solution corresponding to the updated optimal whale individual, and obtain the optimal real-time total path starting spatial coordinates, the optimal number of sub-paths, and the optimal sub-path starting spatial coordinates of each sub-path. Otherwise, perform the next update of the IWOA population. The optimal real-time total path for the epidemic target is obtained based on the optimal spatial coordinates of the starting point of the initial real-time total path, the optimal number of sub-paths, and the optimal spatial coordinates of the starting point of each sub-path.

10. A dual-spectrum visual target monitoring system for infecting plants, used to implement the dual-spectrum visual target monitoring method as described in any one of claims 1-9, characterized in that: The system includes a dual-spectral infected tree target recognition model construction unit, a forest 3D model construction unit, a forest spatial rectangular coordinate system establishment unit, an infected tree target path planning model construction unit, a real-time dual-spectral infected tree image acquisition unit, a dual-spectral infected tree target recognition model application unit, an infected tree target path planning model application unit, and an infected tree target path display unit. These units are connected sequentially. The dual-spectral diseased tree target recognition model building unit is used to acquire historical dual-spectral diseased tree image datasets in different forest scenes, and to construct a dual-spectral diseased tree target recognition model based on the historical dual-spectral diseased tree image datasets using image recognition algorithms. The forest 3D model building unit is used to acquire the 3D data of the target forest and, based on the 3D data, to construct the 3D model of the target forest using 3D modeling algorithms. The forest spatial rectangular coordinate system establishment unit is used to establish a forest spatial rectangular coordinate system based on a target forest 3D model, thereby obtaining a forest 3D model with the forest spatial rectangular coordinate system set; to obtain the spatial location coordinates of all accessible sub-paths in the forest 3D model with the forest spatial rectangular coordinate system set; and to add several monitoring points in the forest 3D model with the forest spatial rectangular coordinate system set, and obtain the monitoring point spatial location coordinates of each monitoring point. The target path planning model construction unit for infected trees is used to construct the target path planning model for infected trees based on the spatial coordinates of the sub-paths of all passable paths and using a swarm intelligence optimization algorithm. The real-time dual-spectral diseased tree image acquisition unit is set at the actual location of the corresponding monitoring point in the forest to collect real-time dual-spectral diseased tree image data from all monitoring points. The dual-spectral diseased tree target recognition model application unit is used to identify diseased trees based on real-time dual-spectral diseased tree image data using the dual-spectral diseased tree target recognition model, and obtain real-time diseased tree target recognition results; it determines whether any real-time diseased tree target recognition result indicates the existence of a diseased tree target, and if any real-time diseased tree target recognition result indicates the existence of a diseased tree target, then the corresponding monitoring point is taken as the diseased tree target point; The application unit of the infected tree target path planning model is used to plan the infected tree target path based on the spatial coordinates of the monitoring points of all infected tree target points and the spatial coordinates of the sub-paths of all passable paths, and obtain the real-time total path of the infected tree target. The Diseased Tree Target Path Display Unit is used to add the real-time total path of the diseased tree target to a forest 3D model with a forest spatial rectangular coordinate system, thereby obtaining a forest 3D model that displays the real-time total path of the diseased tree target.

Images