A method for monitoring the health of artificial forests

Abstract

The present invention relates to the technical field of image data processing, and specifically to a method for monitoring the health of artificial forests, wherein the monitoring method comprises: using unmanned aerial vehicle remote sensing technology to collect artificial forest image data, and constructing the collected image data into multispectral image data to classify the health of artificial forests; constructing an artificial forest health monitoring model based on a deep learning algorithm, and inputting a multispectral image data training set to train the artificial forest health monitoring model; optimizing the artificial forest health monitoring model using a target detection algorithm, and evaluating the health of the artificial forest according to the output result of the optimized artificial forest health monitoring model; the present invention can more accurately monitor and classify the health of artificial forests by combining a graph neural network with a target detection algorithm; using a node classification method of a graph neural network, it can effectively reduce the time and computing resource consumption of model training, while ensuring the efficiency and accuracy of the model.

Description

Technical Field

[0001] The invention relates to the technical field of image data processing, and in particular to a method for monitoring the health of artificial forests. Background Art

[0002] Planted forests refer to forests planted and managed by humans. Compared with natural forests, planted forests have different characteristics in terms of tree species structure, density, and management methods, and therefore have unique requirements for health monitoring.

[0003] At present, plantation health monitoring faces the following challenges:

[0004] At present, conventional image remote sensing technology is used to collect image data, mainly using visible light and near-infrared bands, while ignoring mid-infrared and thermal infrared bands, which are very important for identifying plant moisture content, temperature changes and disease characteristics;

[0005] Plantations are usually composed of a single or a few tree species and are densely packed, making it difficult to distinguish individual trees or identify local health conditions in remote sensing images. In addition, plantations at different growth stages also differ in spectral response, making monitoring more difficult. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for monitoring the health of artificial forests, comprising the following steps:

[0008] The artificial forest image data was collected using UAV remote sensing technology, and the collected image data was constructed into spectral image data to classify the health status of the artificial forest, including:

[0009] Construct a spectral image dataset and perform a preliminary classification of the health status of plantations by calculating the normalized difference vegetation index of plantations.

[0010] A plantation forest health monitoring model is constructed based on a deep learning algorithm, and a spectral image data training set is input to train the plantation forest health monitoring model, including:

[0011] Construct a graph structure based on the spatial relationship of nodes in spectral image data, use graph neural network to build a plantation health monitoring model, and classify plantation health based on the constructed plantation health monitoring model;

[0012] The target detection algorithm is used to optimize the plantation health monitoring model, and the health status is assessed based on the output results of the optimized plantation health monitoring model, including:

[0013] The YOLO target detection algorithm is used to optimize the plantation health monitoring model, specifically:

[0014] The artificial forest health monitoring model is optimized and trained by calculating the positioning loss, confidence loss and classification loss.

[0015] As a preferred solution of the method for monitoring the health of plantations described in the present invention, the initial classification of the health of plantations is specifically as follows:

[0016]

[0017] in, represents the reflectivity of the extracted infrared band, represents the reflectance of the extracted red light band, Represents the calculated normalized difference vegetation index, which is used to make a preliminary classification of the health of plantations, as follows:

[0018] When the calculated normalized difference vegetation index satisfies the formula When , the vegetation in the corresponding artificial forest area in the current image is classified into the healthy category;

[0019] When the calculated normalized difference vegetation index satisfies the formula When , the vegetation in the corresponding artificial forest area in the current image is classified as dead;

[0020] As a preferred solution of the method for monitoring the health of artificial forests described in the present invention, the graph structure constructed based on the spatial relationship of nodes in the multispectral image data set is specifically as follows:

[0021] For two regional nodes , , based on ecological similarity and physical distance, the spatial relationship between the two is described,

[0022]

[0023]

[0024]

[0025] in, Representation Node , The weight of the edge between them indicates the strength of the spatial relationship between nodes. represents the Euclidean distance between nodes, represents the ecological similarity between nodes, represents the environmental impact factor, Represents the distance attenuation factor, which is used to indicate the influence of physical distance on edge weight. It takes a positive value. The size of the attenuation factor is proportional to the influence on edge weight. Represents the ecological similarity adjustment factor, which is used to control the influence of ecological similarity on edge weights. , Respectively represent nodes , of value, Represents the environmental attenuation factor, which is used to represent the difference in environmental conditions between nodes. , Respectively represent nodes , environmental conditions, including light intensity and moisture content.

[0026] As a preferred solution of the method for monitoring the health of an artificial forest described in the present invention, the artificial forest health monitoring model constructed by using a graph neural network is specifically as follows:

[0027] Computational-based The value constructs the feature vector of each node, then we have, , Indicates The initial feature vector of the nodes, Indicates Node value;

[0028] Using the constructed feature vector, we can get the Receive information and update the features, then we have,

[0029] Receive information from neighboring nodes, then we have,

[0030]

[0031] According to the received information, the feature is updated, then,

[0032]

[0033] in, Representation Node No. Layer feature vector, Representation Node , In the The edge weights of the layer, Representation Node , In the Environmental factors of the layer, Represents a message function, which is used to receive information from adjacent nodes. Representation Node From the neighboring nodes The received Layer information, represents a nonlinear activation function, Representation Node In the The updated feature vector of the layer.

[0034] As a preferred solution of the method for monitoring the health of a plantation forest described in the present invention, the plantation forest health classification is specifically as follows:

[0035] The plantation health classification is performed based on the softmax function, specifically:

[0036]

[0037] in, The weight matrix representing the output function is set by the implementer according to the actual application scenario. represents the number of layers of the graph neural network, Representation Node No. Layer feature vector, Represents the softmax function value, which is used to classify the health of plantations. The specific classification is as follows:

[0038] For the current node The probability distribution transformed by the softmax function is based on To classify the health of plantations, represents the largest category, then we have,

[0039] The health degree of artificial forests is divided into three categories: healthy, sub-healthy and unhealthy, and the range of softmax function values ​​for each category is determined;

[0040] For the current node The maximum category in the probability distribution of is selected, and according to the result of the maximum category, the health degree of the artificial forest area corresponding to the current node is matched by category.

[0041] As a preferred solution of the method for monitoring the health of a plantation forest described in the present invention, the optimization of the plantation forest health monitoring model using the YOLO target detection algorithm is specifically as follows:

[0042] For the classification results of the plantation health monitoring model, the remote sensing images corresponding to the areas with different health status are annotated, and the corresponding bounding boxes are generated for healthy, sub-healthy and unhealthy areas;

[0043] The YOLO target detection algorithm is used for optimization, including positioning loss, confidence loss and classification loss. The positioning loss is used to measure the position and size difference between the predicted bounding box and the actual bounding box. The confidence loss is used to measure whether the predicted bounding box contains the target object, and the difference between the predicted confidence and the actual. The classification loss is used to measure the difference between the predicted category probability and the actual category.

[0044] As a preferred solution of the method for monitoring the health of a plantation according to the present invention, the positioning loss is specifically calculated as follows:

[0045] For the loss of center coordinates,

[0046]

[0047] For the loss of width and height,

[0048]

[0049] in, Indicates the size of the network partition. In this embodiment, the size of the network partition is , represents the number of bounding boxes predicted by each network unit, Represents a network unit The bounding box Whether to include the target, represents the center coordinates of the bounding box for tree classification monitored by the plantation health monitoring model, , represents the width and height of the bounding box for tree classification in the plantation health monitoring model, represents the center coordinates of the bounding box predicted by the target detection algorithm for the tree classification, , represents the width and height of the bounding box predicted by the target detection algorithm for tree classification, represents the loss of the center coordinates, Indicates the loss of width and height.

[0050] As a preferred solution of the method for monitoring the health of a plantation described in the present invention, the confidence loss and classification loss are specifically calculated as follows:

[0051]

[0052] in, Indicates the size of the network partition. In this embodiment, the size of the network partition is , represents the number of bounding boxes predicted by each network unit, Represents a network unit The bounding box Whether to include the target, represents the confidence score of tree classification by the plantation health monitoring model, represents the confidence score of the target detection algorithm for the tree classification prediction, represents the confidence loss;

[0053] The classification loss is used to measure the difference between the predicted category probability and the actual category, then,

[0054]

[0055] in, Indicates the size of the network partition. In this embodiment, the size of the network partition is , Represents a network unit The bounding box Whether to include the target, Represents the plantation health monitoring model for trees of category The probability of Indicates that the target detection model treats trees as categories The predicted probability of represents the classification loss.

[0056] Beneficial effects of the present invention: The present invention can more accurately monitor and classify the health of artificial forests by combining graph neural networks and target detection algorithms. This combination method can effectively capture the complex spatial relationships in image data, thereby improving the accuracy of the monitoring model; by fusing the output results of multiple models and algorithms, the artificial forest health monitoring model of the present invention can show stronger robustness in different environments and conditions, reducing the misjudgment or omission that may be caused by a single model; during the training process, the node classification method of the graph neural network and the optimization steps of the target detection algorithm can effectively reduce the time and computing resource consumption of model training, while ensuring the efficiency and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0058] Figure 1 The figure is a schematic diagram of the overall method steps of a method for monitoring the health of an artificial forest according to the present invention.

[0059] Figure 2 This is an effect diagram of plantation classification before optimization of a plantation health monitoring model of a method for monitoring the health level of a plantation in the present invention.

[0060] Figure 3 This is a diagram showing the effect of plantation classification after optimizing and training a plantation health monitoring model using a target detection algorithm in a method for monitoring the health of a plantation in the present invention. DETAILED DESCRIPTION

[0061] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0062] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0063] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0064] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0065] Meanwhile, in the description of the present invention, it should be noted that the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0066] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0067] Example 1

[0068] Reference Figures 1 to 3 , which is the first embodiment of the present invention, provides a method for monitoring the health of a plantation, comprising the following steps:

[0069] S1: Use UAV remote sensing technology to collect plantation forest image data, and construct the collected image data into a multispectral image dataset to classify the health status of plantation forests.

[0070] Specifically, the use of UAV remote sensing technology to collect artificial forest image data is to formulate flight missions for UAVs and use image sensors to collect artificial forest image data. The flight mission is to adjust the spatial resolution and coverage of the image by setting the flight altitude and route. The use of image sensors to collect artificial image data is to use a multispectral camera mounted on the UAV to collect images of the artificial forest area and construct the collected images into a multispectral image dataset. The constructed multispectral image dataset is used for the classification of the health status of artificial forests. The specific construction process is as follows:

[0071] Use a multispectral camera to collect image data of the entire plantation area multiple times, and ensure that there is overlap between multiple images for subsequent image stitching and data integration;

[0072] Before constructing the multispectral image data set, the collected image data is preprocessed. The data preprocessing includes geometric correction of the image data, radiation correction and image stitching to ensure the accuracy of the spectral image data set. The data preprocessing summarized in this embodiment is an existing conventional technical means, which will not be described in detail here.

[0073] For the processed image data , extract the spectral reflectance data of each band from the multi-band image, and construct the extracted spectral reflectance data into a spectral curve, and then calculate the normalized difference vegetation index of the artificial forest, then,

[0074]

[0075] in, represents the reflectivity of the extracted infrared band, represents the reflectance of the extracted red light band, Represents the calculated normalized difference vegetation index, which is used to make a preliminary classification of the health of plantations, as follows:

[0076] When the calculated normalized difference vegetation index satisfies the formula When , the vegetation in the corresponding artificial forest area in the current image is classified into the healthy category;

[0077] When the calculated normalized difference vegetation index satisfies the formula When , the vegetation in the corresponding artificial forest area in the current image is classified as dead;

[0078] It should be noted that for the multispectral image dataset after initial classification, 70% of the health category data was selected as the training set for the subsequent artificial health monitoring model, and 30% of the data set was selected as the test set for the subsequent artificial health monitoring model. The training set was used for model training of the artificial forest health monitoring model to improve the monitoring accuracy of the model, and the test set was used for testing the artificial forest health monitoring model to ensure the accuracy of the artificial forest health monitoring model.

[0079] S2: Build a plantation forest health monitoring model based on deep learning algorithm, and input spectral image data training set to train the plantation forest health monitoring model.

[0080] Specifically, the artificial forest health monitoring model based on deep learning algorithm is to use graph neural network to build the artificial forest health monitoring model. The graph neural network uses the normalized difference vegetation index of different regions in the spectral image dataset as each node of the graph structure, and the spatial relationship between regions as the edge connecting different nodes, thereby constructing a graph structure to achieve spatial interaction and overall health monitoring between artificial forest trees. The specific construction of the artificial forest health monitoring model is as follows:

[0081] Using the multispectral image dataset to construct the graph structure, we have:

[0082] The normalized difference vegetation index of different regions is used as each node of the graph structure. The connection edge between each node is determined according to the spatial relationship between regions. The spatial relationship is described by the spatial distance relationship, specifically:

[0083] For two regional nodes , , based on ecological similarity and physical distance, the spatial relationship between the two is described,

[0084]

[0085]

[0086]

[0087] in, Representation Node , The weight of the edge between them indicates the strength of the spatial relationship between nodes. represents the Euclidean distance between nodes, represents the ecological similarity between nodes, represents the environmental impact factor, Represents the distance attenuation factor, which is used to indicate the influence of physical distance on edge weight. It takes a positive value. The size of the attenuation factor is proportional to the influence on edge weight. Represents the ecological similarity adjustment factor, which is used to control the influence of ecological similarity on edge weights. , Respectively represent nodes , of value, Represents the environmental attenuation factor, which is used to represent the difference in environmental conditions between nodes. , Respectively represent nodes , Environmental conditions, including light intensity and moisture content;

[0088] It should be noted that the distance attenuation factor, environmental attenuation factor, and ecological similarity adjustment factor involved in constructing the graph structure are used in this embodiment to correct the data in the graph construction process to ensure the accuracy of the data. In actual applications, the implementers adjust the values ​​according to the actual application scenarios, and then adjust the degree of correction.

[0089] After the graph structure is built, the artificial forest health monitoring model is constructed through the graph neural network, as follows:

[0090] Computational-based The value constructs the feature vector of each node, then we have, , Indicates The initial feature vector of the nodes, Indicates Node value;

[0091] Using the constructed feature vector, we can get the Receive information and update the features, then we have,

[0092] Receive information from neighboring nodes, then we have,

[0093]

[0094] According to the received information, the feature is updated, then,

[0095]

[0096] in, Representation Node No. Layer feature vector, Representation Node , In the The edge weights of the layer, Representation Node , In the Environmental factors of the layer, Represents a message function, which is used to receive information from adjacent nodes. Representation Node From the neighboring nodes The received Layer information, represents a nonlinear activation function, Representation Node In the The updated feature vector of the layer;

[0097] The plantation health classification is performed based on the softmax function, specifically:

[0098]

[0099] in, The weight matrix representing the output function is set by the implementer according to the actual application scenario. represents the number of layers of the graph neural network, Representation Node No. Layer feature vector, Represents the softmax function value, which is used to classify the health of plantations. The specific classification is as follows:

[0100] For the current node The probability distribution transformed by the softmax function is based on To classify the health of plantations, represents the largest category, then we have,

[0101] The health degree of artificial forests is divided into three categories: healthy, sub-healthy and unhealthy, and the range of softmax function values ​​for each category is determined;

[0102] For the current node The maximum category in the probability distribution of is selected, and according to the result of the maximum category, the health degree of the artificial forest area corresponding to the current node is matched by category.

[0103] S3: Use the target detection algorithm to optimize the plantation forest health monitoring model, and assess the health level based on the output results of the optimized plantation forest health monitoring model.

[0104] Specifically, the use of the target price detection algorithm to optimize the plantation health monitoring model is to use the YOLO target detection algorithm to optimize the training of the plantation health monitoring model, as follows:

[0105] For the classification results of the plantation health monitoring model, the remote sensing images corresponding to the areas with different health status are annotated, and the corresponding bounding boxes are generated for healthy, sub-healthy and unhealthy areas;

[0106] Using the YOLO target detection algorithm for optimization, including positioning loss, confidence loss, and classification loss, we have:

[0107] The positioning loss is used to measure the position and size difference between the predicted bounding box and the actual bounding box, then,

[0108] For the loss of center coordinates,

[0109]

[0110] For the loss of width and height,

[0111]

[0112] in, Indicates the size of the network partition. In this embodiment, the size of the network partition is , represents the number of bounding boxes predicted by each network unit, Represents a network unit The bounding box Whether to include the target, represents the center coordinates of the bounding box for tree classification monitored by the plantation health monitoring model, , represents the width and height of the bounding box for tree classification in the plantation health monitoring model, represents the center coordinates of the bounding box predicted by the target detection algorithm for the tree classification, , represents the width and height of the bounding box predicted by the target detection algorithm for tree classification, represents the loss of the center coordinates, Indicates the loss of width and height;

[0113] Confidence loss is used to measure whether the predicted bounding box contains the target object, and the difference between the predicted confidence and the actual confidence. Then,

[0114]

[0115] in, Indicates the size of the network partition. In this embodiment, the size of the network partition is , represents the number of bounding boxes predicted by each network unit, Represents a network unit The bounding box Whether to include the target, represents the confidence score of tree classification by the plantation health monitoring model, represents the confidence score of the target detection algorithm for the tree classification prediction, represents the confidence loss;

[0116] The classification loss is used to measure the difference between the predicted category probability and the actual category, then,

[0117]

[0118] in, Indicates the size of the network partition. In this embodiment, the size of the network partition is , Represents a network unit The bounding box Whether to include the target, Represents the plantation health monitoring model for trees of category The probability of Indicates that the target detection model treats trees as categories The predicted probability of represents the classification loss;

[0119] According to the optimization results of the YOLO target detection algorithm, the initial monitoring results of the artificial forest health monitoring model were corrected for losses, specifically:

[0120] According to the calculated positioning loss, confidence loss and classification loss, the classification area of ​​the forest is corrected in the artificial forest health monitoring model, and then,

[0121] The calculated positioning loss is used to correct the classification area, the calculated confidence loss is used to correct the precision of the regional bounding box, and the calculated classification loss is used to correct the accuracy of the tree classification;

[0122] Finally, the health status results of all areas are plotted on the map to generate a health status distribution map of the artificial forest, and the generated distribution map is projected on the terminal device in real time so that relevant personnel can monitor the health status of the artificial forest in real time.

[0123] Example 2

[0124] The second embodiment of the present invention provides a monitoring system for the health of plantations, including a plantation health initial classification module, a plantation health monitoring model building module, and a plantation health monitoring model optimization training module;

[0125] Specifically, the artificial forest health status initial classification module is used to initially classify the health status of artificial forests based on the collected spectral image data; the artificial forest health monitoring model building module uses graph neural networks to build an artificial forest health monitoring model; the artificial forest health monitoring model optimization training module uses target detection algorithms to optimize and train the built artificial forest health monitoring model.

[0126] Furthermore, the plantation health status preliminary classification module performs preliminary health status classification of plantations based on the preprocessed spectral image data. Before construction, the collected spectral image data is preprocessed, including denoising, standardization and feature extraction. The processed data is then input into the plantation health status preliminary classification module to ensure the accuracy of classification and the effectiveness of model training. The preliminary classification results are used as input into the plantation health monitoring model construction module to provide preliminary health label data and assist in model training.

[0127] The plantation health monitoring model building module uses graph neural networks to build a plantation health monitoring model, predicts the health status based on the spatial relationship of nodes, and uses the data from the initial classification module for training and generates a preliminary health monitoring model. The generated model will be further optimized;

[0128] The artificial forest health monitoring model optimization training module uses the target detection algorithm to optimize the training of the constructed artificial forest health monitoring model. The prediction accuracy of the model is improved by optimizing the bounding box positioning, classification and confidence of the model. The optimized model is used for the final health assessment. According to the output results of the optimized artificial forest health monitoring model, the health status of the artificial forest is assessed and a distribution map of the health status is generated. Finally, the assessment results are displayed in real time on the terminal device for real-time monitoring by relevant personnel, and a feedback mechanism is provided for dynamic adjustment.

[0129] It should be noted that through the close collaboration between modules, the system functions are more complete, and the front and rear modules are closely linked, forming a closed-loop plantation health monitoring system, which can effectively improve the accuracy and real-time performance of monitoring.

[0130] Furthermore, if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0131] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0132] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk case (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0133] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those features that are not relevant to implementing the invention).

[0134] It should be understood that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for monitoring the health of an artificial forest, characterized in that: The following steps are included: The image data of artificial forests were collected using UAV remote sensing technology, and the collected image data were constructed into spectral image data to classify the health status of artificial forests. include, Construct a spectral image dataset and perform a preliminary classification of the health status of plantations by calculating the normalized difference vegetation index of plantations. The initial classification of the health status of plantations is as follows: in, represents the reflectivity of the extracted infrared band, represents the reflectance of the extracted red light band, Represents the calculated normalized difference vegetation index, which is used to make a preliminary classification of the health of plantations, as follows: When the calculated normalized difference vegetation index satisfies the formula When , the vegetation in the corresponding artificial forest area in the current image is classified into the healthy category; When the calculated normalized difference vegetation index satisfies the formula When , the vegetation in the corresponding artificial forest area in the current image is classified as dead; A plantation forest health monitoring model is constructed based on a deep learning algorithm, and a spectral image data training set is input to train the plantation forest health monitoring model, including: Construct a graph structure based on the spatial relationship of nodes in spectral image data, use graph neural network to build a plantation health monitoring model, and classify plantation health based on the constructed plantation health monitoring model; The graph structure constructed based on the spatial relationship of nodes in the spectral image data is specifically as follows: For two regional nodes , , based on ecological similarity and physical distance, the spatial relationship between the two is described, in, Representation Node , The weight of the edge between them indicates the strength of the spatial relationship between nodes. represents the Euclidean distance between nodes, represents the ecological similarity between nodes, represents the environmental impact factor, Represents the distance attenuation factor, which is used to indicate the influence of physical distance on edge weight. It takes a positive value. The size of the attenuation factor is proportional to the influence on edge weight. Represents the ecological similarity adjustment factor, which is used to control the influence of ecological similarity on edge weights. , Respectively represent nodes , The NDVI value, Represents the environmental attenuation factor, which is used to represent the difference in environmental conditions between nodes. , Respectively represent nodes , Environmental conditions, including light intensity and moisture content; The use of graph neural network to build a plantation health monitoring model is as follows: The feature vector of each node is constructed based on the calculated NDVI value, then, , Indicates The initial feature vector of the nodes, Indicates NDVI value of each node; Using the constructed feature vector, we can get the Receive information and update the features, then we have, Receive information from neighboring nodes, then we have, According to the received information, the feature is updated, then, in, Representation Node No. Layer feature vector, Representation Node , In the The edge weights of the layer, Representation Node , In the Environmental factors of the layer, Represents a message function, which is used to receive information from adjacent nodes. Representation Node From the adjacent nodes The received Layer information, represents a nonlinear activation function, Representation Node In the The updated feature vector of the layer; The target detection algorithm is used to optimize the plantation health monitoring model, and the health status is assessed based on the output results of the optimized plantation health monitoring model, including: The YOLO target detection algorithm is used to optimize the plantation health monitoring model, specifically: The optimization of the plantation health monitoring model using the YOLO target detection algorithm is as follows: For the classification results of the plantation health monitoring model, the remote sensing images corresponding to the areas with different health status are annotated, and the corresponding bounding boxes are generated for healthy, sub-healthy and unhealthy areas; The YOLO target detection algorithm is used for optimization, including positioning loss, confidence loss and classification loss. The positioning loss is used to measure the position and size difference between the predicted bounding box and the actual bounding box. The confidence loss is used to measure whether the predicted bounding box contains the target object and the difference between the predicted confidence and the actual. The classification loss is used to measure the difference between the predicted category probability and the actual category. The artificial forest health monitoring model is optimized and trained by calculating the positioning loss, confidence loss and classification loss.

2. A method for monitoring the health of an artificial forest as claimed in claim 1, characterized in that: The plantation health classification is as follows: The plantation health classification is performed based on the softmax function, specifically: in, The weight matrix representing the output function is set by the implementer according to the actual application scenario. represents the number of layers of the graph neural network, Representation Node No. Layer feature vector, Represents the softmax function value, which is used to classify the health of plantations. The specific classification is as follows: For the current node The probability distribution transformed by the softmax function is based on To classify the health of plantations, represents the largest category, then we have, The health degree of artificial forests is divided into three categories: healthy, sub-healthy and unhealthy, and the range of softmax function values ​​for each category is determined; For the current node The maximum category in the probability distribution of is selected, and according to the result of the maximum category, the health degree of the artificial forest area corresponding to the current node is matched by category.

3. A method for monitoring the health of an artificial forest as claimed in claim 2, characterized in that: The positioning loss is specifically calculated as follows: For the loss of center coordinates, For the loss of width and height, in, Indicates the size of the network partition, represents the number of bounding boxes predicted by each network unit, Represents a network unit The bounding box Whether to include the target, represents the center coordinates of the bounding box for tree classification monitored by the plantation health monitoring model, , represents the width and height of the bounding box for tree classification in the plantation health monitoring model, represents the center coordinates of the bounding box predicted by the target detection algorithm for the tree classification, , represents the width and height of the bounding box predicted by the target detection algorithm for tree classification, represents the loss of the center coordinates, Indicates the loss of width and height.

4. A method for monitoring the health of an artificial forest as claimed in claim 3, characterized in that: The confidence loss and classification loss are specifically calculated as follows: in, Indicates the size of the network partition, represents the number of bounding boxes predicted by each network unit, Represents a network unit The bounding box Whether to include the target, represents the confidence score of tree classification by the plantation health monitoring model, represents the confidence score of the target detection algorithm for the tree classification prediction, represents the confidence loss; The classification loss is used to measure the difference between the predicted category probability and the actual category, then, in, Indicates the size of the network partition, Represents a network unit The bounding box Whether to include the target, Represents the plantation health monitoring model for trees of category The probability of Indicates that the target detection model treats trees as categories The predicted probability of represents the classification loss.

Images