Intelligent garden pest inspection device and method based on deep learning

Abstract

This application provides a deep learning-based intelligent pest inspection device and method for gardens. By constructing a pest identification model within the garden inspection area, pests are identified from image data of vegetation surfaces within the inspection area, yielding pest attribute characteristics and distribution information. Based on environmental parameters and pest distribution information, various pest aggregation areas within the inspection area are determined, and the path relationships between these aggregation areas are established during inspection. Differentiated inspection paths for each aggregation area are determined based on pest attribute characteristics and path relationships. Pest inspection is then conducted using these differentiated paths. Using this approach, the pest inspection paths within the garden inspection area can be dynamically adjusted based on the population characteristics and spatial distribution patterns of pests.

Description

Technical Field

[0001] This application relates to the field of intelligent inspection technology, and more specifically, to an intelligent inspection device and method for garden pests based on deep learning. Background Technology

[0002] Intelligent inspection refers to a modern operating mode that uses advanced technologies such as the Internet of Things, artificial intelligence, robotics, and data analysis to automatically and intelligently monitor and inspect equipment, facilities, or the environment. It replaces the traditional method that relies on manual labor and paper records, greatly improving the efficiency and accuracy of inspections. It is a key technology application to ensure industrial production safety, power grid stability, and efficient urban management.

[0003] In the field of garden pest control, traditional pest inspection mainly relies on manual inspection combined with visual identification. This not only requires a large investment of manpower but also suffers from low inspection efficiency, limited coverage, and strong subjectivity in pest identification, making it difficult to accurately and quickly obtain the population characteristics and spatial distribution of pests in large areas of gardens. Although some automated inspection methods have introduced image acquisition equipment, they mostly use simple image comparison methods for pest identification, resulting in low accuracy. Moreover, the inspection paths are mostly pre-fixed routes without dynamic adjustment based on the actual distribution patterns and growth environment conditions of pests. This leads to insufficient inspection frequency in high-risk pest areas and excessive inspection in low-risk areas, resulting in wasted resources. Therefore, how to dynamically adjust the pest inspection paths in garden inspection areas based on the population characteristics and spatial distribution patterns of pests has become a problem facing the industry. Summary of the Invention

[0004] This application provides a deep learning-based intelligent inspection device and method for garden pests, which can dynamically adjust the inspection path of pests in the garden inspection area based on the population characteristics and spatial distribution patterns of pests in the garden inspection area.

[0005] Firstly, this application provides a deep learning-based intelligent inspection method for garden pests, comprising the following steps: Collect image data of vegetation surfaces within the garden inspection area; A pest identification model for the garden inspection area is constructed based on deep learning. Based on the pest identification model, pest identification is performed on the image data to obtain the pest attribute characteristics and pest distribution information on the vegetation in the garden inspection area. The environmental parameters within the garden inspection area are monitored, and the pest aggregation is divided into different areas based on the environmental parameters and the pest distribution information. Each aggregation area of ​​pests in the garden inspection area is obtained. The path association between each aggregation area is determined based on deep learning and the preset pest inspection path in the garden inspection area. Based on the pest attributes and the path relationships between various clusters, the preset pest inspection paths within the garden inspection area are adjusted to obtain differentiated inspection paths for each cluster. Pest inspections are conducted in the garden inspection area using differentiated inspection routes for each cluster area.

[0006] In some embodiments, constructing a pest identification model within a garden inspection area based on deep learning specifically includes: Acquire historical pest image data within the garden inspection area; The historical pest image data is divided into training set, validation set and test set to construct a pest identification model in the garden inspection area based on deep learning.

[0007] In some embodiments, pest identification based on the pest identification model in the image data to obtain pest attribute characteristics and pest distribution information on vegetation within the garden inspection area specifically includes: The image data is loaded into the pest identification model, and pest identification is performed on each image in the image data to obtain pest information for each image. Based on all pest information, determine the pest characteristics and distribution information of the vegetation in the garden inspection area.

[0008] In some embodiments, the garden inspection area is divided into pest aggregation zones based on the environmental parameters and the pest distribution information, resulting in various pest aggregation zones within the garden inspection area, specifically including: The environmental parameters are standardized and preprocessed to obtain the preprocessed environmental parameters; The criteria for classifying pest clusters in the garden inspection area are determined based on the environmental parameters and the pest distribution information. Based on the aforementioned criteria for classifying pest clusters, the garden inspection area is divided to obtain various clusters of pests within the garden inspection area.

[0009] In some embodiments, determining the path association between various clustered areas during pest inspection based on deep learning and pre-defined pest inspection paths within the garden inspection area specifically includes: Obtain the preset pest inspection path within the garden inspection area; Determine the intersection characteristics of the preset pest inspection path with each gathering area during the pest inspection process; Based on deep learning and the aforementioned intersection features, the path associations between various aggregation areas are determined during pest inspection.

[0010] In some embodiments, the preset pest inspection paths within the garden inspection area are differentiated based on the pest attribute characteristics and the path association relationships between various cluster areas, resulting in differentiated inspection paths for each cluster area, specifically including: Based on the characteristics of the pests, determine the inspection schedule information for each cluster area; Based on the path relationships between various cluster areas, determine the path differentiation adjustment information within the garden inspection area; By using the inspection arrangement information and the path differentiation adjustment information, the preset pest inspection paths within the garden inspection area are differentiated to obtain differentiated inspection paths for each cluster area.

[0011] In some embodiments, conducting pest inspections in garden inspection areas using differentiated inspection paths for various clustered areas specifically includes: Input the differentiated inspection paths of each cluster area into the inspection equipment of the garden inspection area; The inspection equipment performs pest inspections on the garden inspection area according to the differentiated inspection paths of each cluster area.

[0012] Secondly, this application provides a deep learning-based intelligent inspection device for garden pests, comprising: The data acquisition module is used to collect image data of the vegetation surface within the garden inspection area; The processing module is used to construct a pest identification model in the garden inspection area based on deep learning, and to identify pests in the image data based on the pest identification model to obtain the pest attribute characteristics and pest distribution information on the vegetation in the garden inspection area. The processing module is also used to monitor environmental parameters within the garden inspection area, divide the garden inspection area into pest clusters based on the environmental parameters and the pest distribution information, obtain various clusters of pests within the garden inspection area, and determine the path association between various clusters during the pest inspection process based on deep learning and the preset pest inspection path within the garden inspection area. The processing module is also used to differentiate the preset pest inspection path in the garden inspection area based on the pest attribute characteristics and the path association between each cluster area, so as to obtain the differentiated inspection path for each cluster area. The execution module is used to conduct pest inspections in the garden inspection area through differentiated inspection paths for each cluster area.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described deep learning-based intelligent inspection method for garden pests.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned deep learning-based intelligent inspection method for garden pests.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The intelligent pest inspection device and method for gardens based on deep learning provided in this application first collects image data of vegetation surfaces within a garden inspection area; constructs a pest identification model within the garden inspection area based on deep learning, and identifies pests from the image data based on the pest identification model to obtain pest attribute characteristics and distribution information on the vegetation within the garden inspection area; monitors environmental parameters within the garden inspection area, and divides the garden inspection area into pest clusters based on the environmental parameters and the pest distribution information to obtain various clusters of pests within the garden inspection area; determines the path association relationship between various clusters during the pest inspection process based on deep learning and a preset pest inspection path within the garden inspection area; differentiates the preset pest inspection path within the garden inspection area based on the pest attribute characteristics and the path association relationship between various clusters to obtain differentiated inspection paths for each cluster; and performs pest inspection in the garden inspection area using the differentiated inspection paths for each cluster.

[0016] Therefore, this application, in the process of intelligent inspection of garden pests, collects image data of vegetation surfaces, constructs a pest identification model based on deep learning, and performs image recognition on the image data. This obtains the attribute characteristics and basic distribution information of pests within the garden, providing core data for analyzing the spatial distribution patterns of pests and adjusting paths based on population characteristics. By monitoring environmental parameters and combining them with pest distribution information to delineate pest gathering areas, and relying on deep learning to determine the path relationships within these gathering areas, the spatial distribution characteristics of pests are refined, and a path framework for these gathering areas is established, enabling adjustments to the inspection path. This method can accurately reflect the actual spatial distribution patterns of pests. It adjusts preset inspection paths based on pest attributes and the path relationships within aggregation zones, achieving dynamic adjustment of inspection paths that deeply integrates pest population characteristics with spatial distribution patterns. This allows the paths to adapt to the pest characteristics and spatial relationships in different aggregation areas. Actual inspections are conducted based on these differentiated inspection paths for each aggregation area, ensuring that the adjusted paths, based on pest population characteristics and spatial distribution patterns, are implemented in actual inspection work. This allows inspection operations to precisely match the population and spatial distribution characteristics of pests, improving the targeting and efficiency of garden pest inspections. Using this solution, pest inspection paths in garden inspection areas can be dynamically adjusted based on the population characteristics and spatial distribution patterns of pests within the inspection area. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a deep learning-based intelligent inspection method for garden pests, as shown in some embodiments of this application. Figure 2 This is an exemplary flowchart illustrating the determination of a clustered region according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of differentiated inspection paths according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a deep learning-based intelligent inspection device for garden pests, as shown in some embodiments of this application. Figure 5 This is a schematic diagram of the structure of a computer device that implements a deep learning-based intelligent inspection method for garden pests, according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a deep learning-based intelligent inspection method for garden pests, according to some embodiments of this application. This deep learning-based intelligent inspection method for garden pests mainly includes the following steps: In step 101, image data of the vegetation surface within the garden inspection area is collected.

[0020] It should be noted that the image data in this application represents a digital carrier of visual information about the surface of vegetation within the garden inspection area. It reflects the growth status of different types of vegetation such as trees, shrubs, and herbs in the area, as well as the actual situation of whether pests are attached to the vegetation surface, the attachment location of the pests, and their morphological characteristics. This includes high-definition original images of key parts of leaves, branches, flowers, and fruits of various vegetation types, geographic coordinates and timestamp metadata corresponding to the image acquisition, and standardized images after preprocessing. Specifically, the preprocessing operation involves using a 3×3 mean filter convolution kernel to perform convolution operations on the image data to eliminate random noise interference, and using a bilinear interpolation algorithm to uniformly adjust all images to a fixed size of 224×224 pixels, while retaining metadata such as geographic coordinates, acquisition time, and vegetation type corresponding to each image acquisition. Preferably, in some embodiments, a mobile inspection device equipped with a high-definition image acquisition module or a fixedly deployed image monitoring node is used to perform full-coverage image acquisition of key parts of leaves, branches, flowers, and fruits of different types of vegetation such as trees, shrubs, and herbs within the garden inspection area.

[0021] In step 102, a pest identification model for the garden inspection area is constructed based on deep learning. Based on the pest identification model, pest identification is performed on the image data to obtain the pest attribute characteristics and pest distribution information on the vegetation in the garden inspection area.

[0022] In some embodiments, constructing a pest identification model within a garden inspection area based on deep learning can be achieved through the following steps: Acquire historical pest image data within the garden inspection area; The historical pest image data is divided into training set, validation set and test set to construct a pest identification model in the garden inspection area based on deep learning.

[0023] In practice, the process begins by collecting pest image archives accumulated from past inspections by the garden management department and publicly shared garden pest image resources from the plant protection station. Simultaneously, supplementary images of common pests and their corresponding host vegetation within the inspection area are collected through manual on-site photography, thus constructing a historical pest image database for the garden inspection area. This historical pest image data is then preprocessed, using mean filtering to remove random noise. Bilinear interpolation is used to uniformly adjust all images to a fixed pixel size. Each image is manually labeled with the pest species, life stage, and attachment location. After labeling, two people cross-check the labels to ensure accuracy. To ensure accuracy, invalid images, including blurry ones and those with incorrect annotations, were removed. Then, the preprocessed historical pest image data was randomly divided into training, validation, and test sets in a 7:2:1 ratio. During this division, the proportion of images of different pest species and life stages in each set was ensured to be consistent with the original dataset to avoid data distribution bias affecting model building performance. The model was built using a convolutional neural network architecture from deep learning. The construction process sequentially included an input layer, convolutional layer, pooling layer, fully connected layer, and output layer. The input layer received preprocessed images of a fixed size, and the convolutional layers used 3×3 convolutional layers. The kernel extracts features with a stride of 1. After each convolutional layer, a 2×2 window max pooling layer is connected to compress the feature dimension and retain key features. This combination of multiple convolutional and pooling layers achieves deep feature mining of the image. Fully connected layers map the extracted features to fixed-dimensional feature vectors. The output layer uses a softmax classification function to output the probability value of the corresponding pest category. During training, the training set is used as input, and gradient descent is used to iteratively optimize the model parameters. Cross-entropy is used as the loss function to measure the deviation between the prediction and labeling results. After each training round, the model performance is tested using a validation set. Yes, the number of convolutional kernels, learning rate, and number of iterations can be adjusted based on the recognition accuracy of the validation set. Training is stopped when the validation set accuracy stabilizes and no longer improves. Finally, the trained model is evaluated using the test set, and the recognition accuracy, recall, and F1 score are calculated. If the evaluation metrics do not meet the preset standards, the data preprocessing method or model structure parameters are readjusted, and the above training process is repeated until the model meets the requirements for identifying garden pests. This model is then used as the pest identification model for identifying pests in the garden inspection area. Other methods can be used in other embodiments, which are not limited here.

[0024] It should be noted that the historical pest image data in this application represents a collection of digitized images accumulated in the past of the garden inspection area, containing different pest species, ages, attachment sites, and corresponding host vegetation backgrounds. It reflects the morphological characteristics of pests at different growth stages, the attachment relationship between pests and vegetation, and the visual presentation characteristics of pests under different environmental conditions. The pest identification model represents a garden pest identification model built on a multi-layer network structure, which has the ability to extract and classify image features. It reflects the model's learning and fitting ability and classification accuracy of garden pest target features. It can be used to identify pests in newly collected garden vegetation images and output feature information such as pest species, quantity, and distribution.

[0025] In some embodiments, the following steps can be used to identify pests in the image data based on the pest identification model to obtain the attribute characteristics and distribution information of pests on vegetation within the garden inspection area: The image data is loaded into the pest identification model, and pest identification is performed on each image in the image data to obtain pest information for each image. Based on all pest information, determine the pest characteristics and distribution information of the vegetation in the garden inspection area.

[0026] In specific implementation, the image data is loaded into the pest identification model, and pest identification is performed on each image in the image data to obtain pest information for each image. This can be achieved in the following way: the image data of garden vegetation is loaded into the trained pest identification model. The pest identification model identification stage is as follows: first, the input layer receives the image data, and then features are extracted through a structure of three convolutional layers and three pooling layers connected alternately. The convolutional layers use 3×3 convolutional kernels, and convolution calculation is performed with a stride of 1 and same padding. The number of convolutional kernels is set to 32, 64, and 128 respectively. The feature mining process proceeds from shallow to deep layers. The pooling layer employs a 2×2 window max pooling operation, compressing the feature dimension with a stride of 2 while retaining key feature information. Then, two fully connected layers map the extracted deep features into feature vectors of fixed dimensions. The number of neurons in the fully connected layers is set to 1024 and 512, respectively. Finally, the softmax classification function in the output layer completes the discrimination, outputting the pest information corresponding to each image, specifically including pest species, age level, number of individuals, attachment site on vegetation, and degree of damage. Other methods can be used in other embodiments, which are not limited here.

[0027] In addition, in specific implementation, determining the pest attributes and distribution information of vegetation within the garden inspection area based on all pest information can be achieved in the following way: First, the pest information of all images is matched one-to-one with its corresponding geographic coordinate metadata. Then, according to the preset 10m×10m spatial grid division rule for the garden inspection area, the pest information of each coordinate point is entered into the corresponding grid cell. For blank grid cells without image acquisition, Kriging interpolation is used to complete the spatial data. The interpolation calculation is then used to predict the pest density and species distribution within the blank grid cells. Finally, the data of all grid cells in the entire area is analyzed. The pest data is summarized and statistically analyzed, including the pest population density, the proportion of dominant pest species, the proportion of different insect age groups, and the proportion of infested area in each grid unit. This is used to determine the pest attributes and characteristics on the vegetation in the garden inspection area. At the same time, by drawing a heat map of pest species distribution and generating a spatial distribution coordinate data table of pest density, the K-means clustering algorithm is used to divide the pest aggregation core area and the diffusion edge area. The spatial distribution range, aggregation degree, and diffusion trend of different pest species are analyzed to obtain complete and accurate pest distribution information. Other methods can be used in other embodiments, which are not limited here.

[0028] It should be noted that the pest information in this application reflects the real-time status of pests on vegetation within the coverage area of ​​a single image, and can be used as the original data source for extracting pest attribute features and analyzing pest distribution information; the pest attribute features reflect the core characteristics of the population composition, age structure, dominant species and degree of damage of pests in the garden inspection area, and can be used as the core basis for classifying the risk categories of pest areas; the pest distribution information reflects the spatial aggregation location, distribution range and diffusion trend of pests in the garden inspection area, and can be used to support the differentiated adjustment of inspection paths and the formulation of precise prevention and control strategies.

[0029] In step 103, environmental parameters within the garden inspection area are monitored, and pest aggregation is divided into different areas based on the environmental parameters and the pest distribution information. Each aggregation area of ​​pests within the garden inspection area is obtained, and the path association between the aggregation areas is determined based on deep learning and the preset pest inspection path within the garden inspection area.

[0030] It should be noted that the environmental parameters in this application represent the quantitative monitoring data of various ecological factors affecting the survival, reproduction, and activity patterns of pests within the garden inspection area. They reflect the adaptability of the ecological environment conditions in the area to the growth, development, population spread, and degree of damage of different types of pests. These parameters include air temperature, relative humidity, light intensity, soil moisture content, soil pH, air circulation speed, as well as the collection time and geographical coordinates corresponding to each parameter. They can be used in conjunction with pest distribution information to classify pest regional risk categories and provide an environmental adaptability basis for judging the probability of pest outbreaks in different areas and optimizing inspection routes.

[0031] In some embodiments, reference Figure 2 The figure is an exemplary flowchart for determining the aggregation area in some embodiments of this application. In this embodiment, the pest aggregation area of ​​the garden inspection area is divided according to the environmental parameters and the pest distribution information. The following steps can be used to obtain the various aggregation areas of pests in the garden inspection area: In step 1031, the environmental parameters are standardized and preprocessed to obtain preprocessed environmental parameters; In step 1032, the criteria for dividing the pest clusters in the garden inspection area are determined based on the environmental parameters and the pest distribution information. In step 1033, the garden inspection area is divided based on the pest aggregation classification criteria to obtain various pest aggregation areas in the garden inspection area.

[0032] In specific implementation, the environmental parameters are standardized and preprocessed. The preprocessed environmental parameters can be obtained in the following way: the environmental parameters collected in the garden inspection area are standardized and preprocessed using the minimum-maximum normalization method. During the operation, the maximum and minimum values ​​of each environmental parameter are first counted among all monitoring points in the entire inspection area. Then, the corresponding parameter values ​​of each monitoring point are adjusted to the range of 0 to 1 to eliminate the numerical differences caused by different units of measurement of different environmental parameters. At the same time, the three-standard-deviation principle is used to identify and remove abnormal data. Specifically, the average and standard deviation of all monitoring values ​​of each environmental parameter are calculated. Values ​​that exceed the range of the average minus three standard deviations to the average plus three standard deviations are judged as abnormal data and removed. Finally, the preprocessed environmental parameters are obtained. Other methods can be used in other embodiments, which are not limited here.

[0033] In addition, in specific implementation, the basis for determining the pest aggregation division of the garden inspection area based on the environmental parameters and the pest distribution information can be implemented in the following way: the entire garden inspection area is divided into several spatial grid units according to the specification of 10 meters by 10 meters, and a multi-dimensional feature vector is constructed for each grid unit. This feature vector covers two core contents: first, the preprocessed environmental parameters, and second, the quantified pest distribution information within the grid unit, specifically including the normalized value of pest population density, the proportion of dominant pest species in the total number of pests, and the proportions of adult and nymph numbers in the total number of pests, etc. This multi-dimensional feature vector is used as the basis for pest aggregation division. Other methods can also be used in other embodiments, which are not limited here.

[0034] In addition, in specific implementation, the division of the garden inspection area based on the pest aggregation classification criteria can be achieved in the following way: A clustering algorithm is used for area division. First, based on past experience in garden pest control, the number of cluster categories is preset, generally set to three categories, corresponding to high-risk, medium-risk, and low-risk pest areas respectively. Based on the pest aggregation classification criteria, i.e., each grid cell encompasses preprocessed environmental parameters, normalized pest population density values, the proportion of dominant pest species, and the proportion of adult and nymph numbers, a multi-dimensional feature vector is randomly selected from all grid cells, with the same number of categories as preset. These grid cells' multi-dimensional feature vectors are used as the initial cluster centers. Next, the feature vectors of all other grid cells are calculated and compared with those of each cluster. The spatial distance is calculated by taking the square root of the sum of squares of the numerical differences in each feature dimension and assigning each grid cell to the nearest cluster. Then, the average value of each indicator in the feature vector of all grid cells within each cluster is calculated, and the cluster center of that cluster is updated using this average value. This process of calculating distance, assigning categories, and updating cluster centers is repeated until the position of the cluster center no longer changes, or until a preset maximum number of iterations is reached (usually one hundred). Finally, grid cells that are spatially adjacent and belong to the same cluster are merged to form contiguous areas, resulting in various pest aggregation areas in the garden inspection area. Each aggregation area is labeled with corresponding environmental parameter features and pest distribution. Other methods can be used in other embodiments, which are not limited here.

[0035] It should be noted that the pest clustering criteria in this application represent the basis for dividing pest clusters within the garden inspection area, reflecting the degree of adaptability of environmental conditions within each grid unit to the survival and reproduction of pests and the overall risk level of the pest population; clustering areas represent risk control units of contiguous pests within the garden inspection area, reflecting the consistency of environmental parameters and pest distribution characteristics within the same area and the degree of risk difference between different areas.

[0036] In some embodiments, determining the path relationships between various clustering areas during pest inspection based on deep learning and preset pest inspection paths within the garden inspection area can be achieved through the following steps: Obtain the preset pest inspection path within the garden inspection area; Determine the intersection characteristics of the preset pest inspection path with each gathering area during the pest inspection process; Based on deep learning and the aforementioned intersection features, the path associations between various aggregation areas are determined during pest inspection.

[0037] It should be noted that the preset pest inspection path in this application is a digital carrier of the pest inspection operation route pre-planned based on the topography, vegetation distribution pattern, infrastructure layout and past pest inspection experience of the garden inspection area. It reflects the path planning idea of ​​full coverage of the area, equipment accessibility and balanced inspection efficiency in the routine garden inspection scenario. It includes the continuous spatial coordinate sequence of the path, preset fixed inspection nodes, prescribed inspection travel direction, suitable inspection equipment type, effective travel time of the path, location marking of impassable obstacles and corresponding alternative detour paths.

[0038] In specific implementation, determining the intersection characteristics of the preset pest inspection path with each cluster area during the pest inspection process can be achieved in the following way: In the geographic information system, the preset pest inspection path layer is overlaid and compared with the cluster area layer. The preset path is traversed segment by segment, and the intersection nodes of the path with each cluster area are identified one by one. The entrance coordinates, exit coordinates, and corresponding directions of each cluster area on the preset path are determined. At the same time, the actual length of the preset pest inspection path passing through each cluster area, the number and spacing of fixed inspection nodes passed by the path in the area are counted. Obstacles along the path in the area are carefully investigated and classified into impassable obstacles and detourable obstacles. The coordinates, range, path offset distance required for detour, and time increment of the obstacle points are recorded. The intersection characteristics of each cluster area with the preset path are comprehensively formed, covering the intersection location, path coverage length, node distribution, passage constraints, and obstacle detour scheme. Other methods can also be used in other embodiments, which are not limited here.

[0039] In addition, in specific implementation, the path association relationship between various aggregation areas during pest inspection can be determined based on deep learning and the aforementioned intersection features in the following way: First, a basic convolutional neural network in deep learning is used to construct an intersection feature screening model. Historical measured data from the garden inspection field are selected as training samples. The sample set covers the original data of intersection features of various areas in past garden pest inspections (including intersection node coordinates, path coverage length, obstacle point distribution location, path passage constraints, etc.), measured data of path association parameters corresponding to the intersection features, and abnormal record data from inspection operations. The sample set is preprocessed by cleaning the data to remove missing values ​​and extreme outliers. The minimum-maximum normalization method is used to uniformly map intersection feature data of different dimensions to the 0-1 interval, completing the dimensional unification and standardization of the sample features. Then, the preprocessed data is further processed... The processed samples were randomly divided into training and validation sets in a 7:3 ratio. A simple structural configuration was applied to the basic convolutional neural network. The normalized intersection feature vector was set as the network input layer, and 2-3 fully connected layers were set in the hidden layers to complete the nonlinear mapping of features. The output layer had two output branches: one was a binary classification output to determine whether the input features were valid or invalid, and the other was a numerical output to output the quantized optimized weights of the valid features. During training, gradient descent was used to iteratively optimize the cross-entropy loss function of the network. The training set drove the model to learn the intrinsic correlation between intersection features and path association parameters. At the same time, the feature selection accuracy and weight optimization precision of the model were verified in real time using the validation set. When the feature selection accuracy of the model on the validation set reached more than 95%, and the loss function value converged to the preset value of 0 for multiple consecutive iterations, the model was considered successful.When the threshold 001 stops decreasing, model training is stopped, resulting in a convergent and stable intersection feature selection model suitable for garden pest inspection scenarios. The intersection features of each cluster area are input into this model, which automatically filters out invalid features unrelated to path association, such as redundant point information that does not intersect the path. Effective intersection features are then quantitatively optimized to improve accuracy. Based on the selected intersection features of each cluster area, the connectivity between any two cluster areas via a preset pest inspection path is analyzed. Combining this with the rated moving speed of the corresponding inspection equipment, and deducting obstacle detour time and node dwell monitoring time, the actual path length, estimated travel time, number of transfer nodes, and connection smoothness are accurately calculated between areas. Simultaneously, the calculated correlation parameters are simply corrected using the trained intersection feature selection model, eliminating outliers to ensure accurate alignment with the actual inspection scenario. Differential path association priorities are assigned based on the risk level of the cluster areas (high, medium, low), with path associations between high-risk areas having the highest priority. To ensure smooth connectivity, traffic efficiency can be appropriately considered in low- and medium-risk areas. The order of inspections and the priority of paths between areas should be clearly defined. Each clustered area is treated as a network node, and the path association parameters, priorities, and detour routes between areas are used as connection edge attributes to construct a complete path association network. The necessary path segments, alternative path segments, and emergency detour routes between areas are clearly marked. Through full-link connectivity verification, each clustered area is checked to ensure it can connect with other areas via associated paths without any traffic interruptions. For path segments with connectivity defects, temporary connection paths are added or the path direction is adjusted based on terrain conditions. Ultimately, the inspection and traffic connection rules and relationship network, which integrates core attributes such as path length, estimated travel time, number of transfer nodes, obstacle detour routes, inspection priority, and equipment compatibility type, is determined as the path association relationship between various clustered areas during pest inspections. Other methods can be used in other embodiments, which are not limited here.

[0040] It should be noted that the intersection feature in this application represents the spatial connection information obtained after spatial overlay analysis of the preset pest inspection path and each pest gathering area, reflecting the spatial adaptability of the preset path and each gathering area, the passage constraints, and the obstacle response requirements; the path association relationship represents the inspection passage connection rules and relationship network between each gathering area in the garden inspection area, reflecting the inspection connection method, passage efficiency, and priority differences between each gathering area, which can be used to optimize and adjust the differentiated inspection path in the garden inspection area and to efficiently execute the inspection operation.

[0041] In step 104, based on the pest attribute characteristics and the path association between each cluster area, the preset pest inspection path in the garden inspection area is adjusted to obtain a differentiated inspection path for each cluster area.

[0042] In some embodiments, reference Figure 3 The figure is an exemplary flowchart for determining differentiated inspection paths in some embodiments of this application. In this embodiment, the preset pest inspection paths within the garden inspection area are differentiated based on the pest attribute characteristics and the path associations between various cluster areas. The differentiated inspection paths for each cluster area can be obtained by the following steps: In step 1041, the inspection schedule information for each cluster area is determined based on the pest attribute characteristics; In step 1042, path differentiation adjustment information within the garden inspection area is determined based on the path association relationships between various cluster areas; In step 1043, the preset pest inspection paths within the garden inspection area are adjusted using the inspection arrangement information and the path differentiation adjustment information to obtain differentiated inspection paths for each cluster area.

[0043] In practice, determining the inspection schedule information for each cluster area based on the pest attribute characteristics can be achieved in the following way: Extract characteristic indicators such as pest population density, dominant pest species, insect age structure, and the proportion of vegetation damage within each cluster area from the pest attribute characteristics. Assign specific quantitative weights to each indicator using a weighting method. Based on historical experience in garden pest control, classify the inspection levels as high, medium, and low. For high-level cluster areas, increase the inspection frequency from once per day to twice per day and extend the inspection dwell time in each area. The inspection requirements include extending the inspection time to 1.5 times the original duration, adding key monitoring sub-nodes for pest aggregation points and severely damaged vegetation patches; maintaining the regular inspection frequency and duration for medium-level aggregation areas, and retaining the original core monitoring nodes; and reducing the inspection frequency from once a day to once every three days for low-level aggregation areas, and merging inspection tasks for adjacent areas of the same level. This forms an inspection arrangement information that includes inspection level, frequency, duration, and key monitoring points. Other methods can be used in other embodiments, which are not limited here.

[0044] In addition, in specific implementation, the path differentiation adjustment information within the garden inspection area can be determined based on the path association between various clustered areas in the following way: Based on the path association between various clustered areas, including the necessary path segments, alternative path segments, obstacle detour schemes, and path priority ranking between areas, a spatial path optimization method is adopted. For paths between high-inspection-level areas, the necessary path segments with high traffic efficiency and ≤2 obstacle points are selected first. At the same time, the connection order of transfer nodes is optimized to reduce the number of path turns and reduce the time consumption. For paths between medium and low-inspection-level areas, alternative path segments that are more than 10% shorter than the original path distance are selected, and redundant path segments with a length <50 meters and no monitoring nodes are eliminated from the path association. For path segments with impassable obstacles such as water bodies and steep slopes, temporary connecting paths are planned in conjunction with obstacle detour schemes, and the detour distance is controlled to not exceed 1.2 times the original path length. This forms path differentiation adjustment information that includes the preferred path scheme, the redundant path elimination list, and the temporary detour path planning. Other methods can also be used in other embodiments, which are not limited here.

[0045] In addition, in specific implementation, the pre-set pest inspection paths within the garden inspection area are differentiated by using the inspection arrangement information and the path differentiation adjustment information. The differentiated inspection paths for each cluster area can be achieved in the following way: Using a path topology correction method, the inspection level requirements in the inspection arrangement information are matched and integrated with the path differentiation adjustment information. In high-inspection-level cluster areas, new inspection sub-nodes such as core pest clusters and severely damaged vegetation areas are added, extending the path segment to within 5 meters of the area edge to achieve full area coverage. In low-inspection-level cluster areas, redundant monitoring nodes and invalid path segments with no abnormal records in three consecutive inspections are deleted, and adjacent short path segments are merged. To improve traffic efficiency, after the path adjustment is completed, the connectivity of the adjusted path is checked using a spatial topology verification method. The connection sequence of nodes in each cluster area is checked one by one, and whether there are any breaks or loops in the path. This ensures that all cluster areas are covered and there are no duplicate inspection path segments. At the same time, the total inspection time, equipment energy consumption, node coverage density and other core indicators of the adjusted path are calculated and compared with the operating parameters of the preset pest inspection path. If the total time after adjustment exceeds the preset value by 5%, the node connection sequence is optimized. If the energy saving ratio is less than 8%, a better alternative path is selected. After local fine-tuning, the differentiated inspection path for each cluster area is finally obtained. Other methods can also be used in other embodiments, which are not limited here.

[0046] It should be noted that the inspection arrangement information in this application reflects the inspection needs of different cluster areas based on the characteristics of pest attributes. This information can be used to clarify the inspection operation standards of each area in the garden pest inspection area and guide the work intensity and key directions of inspection personnel or equipment. The path differentiation adjustment information reflects the optimization direction and specific plan of the preset pest inspection path. This information can be used to provide accurate basis for adjusting the preset path in the garden pest inspection area and ensure the scientific and operable nature of the path adjustment. The differentiated inspection path reflects an efficient and accurate inspection route plan that adapts to the inspection needs of each cluster area. This information can be used to support the efficient implementation of inspection operations in the garden pest inspection area, reduce inspection time and equipment energy consumption, and improve the targeting and efficiency of pest monitoring.

[0047] In step 105, pest inspection is carried out in the garden inspection area through differentiated inspection paths for each cluster area.

[0048] In some embodiments, pest inspection of garden inspection areas using differentiated inspection paths for each cluster area can be achieved through the following steps: Input the differentiated inspection paths of each cluster area into the inspection equipment of the garden inspection area; The inspection equipment performs pest inspections on the garden inspection area according to the differentiated inspection paths of each cluster area.

[0049] In practice, the differentiated inspection paths for each cluster area are first converted into a vector data format that the inspection equipment can recognize. This involves converting the continuous spatial coordinate sequence of the path segments in each cluster area, the location of key monitoring nodes, the inspection direction, and obstacle detour plans into a vector data format that the inspection equipment can read. Simultaneously, the accompanying inspection schedule information is encapsulated into a parameter configuration file that the equipment can read. Then, the converted path vector data and parameter configuration file are input to the corresponding inspection equipment via wired transmission or wireless LAN communication. A georeferencing method is used to calibrate the path coordinates with the positioning module of the inspection equipment. By comparing the coordinate deviations of landmark terrain features within the park, the path parameters are fine-tuned to ensure that the path data perfectly matches the actual geographic space of the park. After the inspection equipment starts operation, it first loads the differentiated inspection path data and parameter configuration file, and then uses the positioning module to obtain its own location information in real time and compare it with the path coordinates. The system matches and executes inspection tasks starting from high-risk cluster areas according to preset inspection priorities. During its movement, it uses lidar or visual sensors to perceive obstacles ahead in real time. If it matches a preset obstacle point, it automatically activates a detour plan to adjust its trajectory. When the device reaches a key monitoring node in the path, it automatically triggers a stop command and activates the image acquisition module and environmental sensing module to collect and record pest images, population numbers, age structure, and real-time environmental parameters. After completing the inspection task of one cluster area, the device automatically switches to the path segment of the next cluster area and repeats the operations of positioning and matching, path movement, obstacle avoidance, and node monitoring until it covers the differentiated inspection paths of all cluster areas. After the inspection operation is completed, the device automatically summarizes all collected data and path execution status, generates an inspection operation report, and uploads it to the back-end management system, thereby completing the pest inspection work in the garden inspection area based on differentiated inspection paths.

[0050] In another aspect, in some embodiments, this application provides a deep learning-based intelligent inspection device for garden pests, referencing... Figure 4 The figure is a schematic diagram of the structure of a deep learning-based intelligent inspection device for garden pests according to some embodiments of this application. The deep learning-based intelligent inspection device 400 for garden pests includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire image data of the vegetation surface within the garden inspection area; Processing module 402, in this application, is used to construct a pest identification model in the garden inspection area based on deep learning, and to identify pests in the image data based on the pest identification model, so as to obtain the pest attribute characteristics and pest distribution information on the vegetation in the garden inspection area. It should be noted that the processing module 402 in this application is also used to monitor environmental parameters within the garden inspection area, divide the garden inspection area into pest clusters based on the environmental parameters and the pest distribution information, obtain each cluster of pests in the garden inspection area, and determine the path association relationship between each cluster of pests during the pest inspection process based on deep learning and the preset pest inspection path within the garden inspection area. In addition, it should be noted that the processing module 402 in this application is also used to differentiate the preset pest inspection path in the garden inspection area based on the pest attribute characteristics and the path association relationship between each cluster area, so as to obtain the differentiated inspection path of each cluster area. The execution module 403 in this application is mainly used to conduct pest inspections in the garden inspection area through differentiated inspection paths of various cluster areas.

[0051] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described deep learning-based intelligent inspection method for garden pests.

[0052] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a deep learning-based intelligent inspection method for garden pests, according to some embodiments of this application. The deep learning-based intelligent inspection method for garden pests in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0053] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0054] The communication bus 502 can be used to transmit information between the aforementioned components.

[0055] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0056] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The method used in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0057] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0058] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0059] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0060] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described deep learning-based intelligent inspection method for garden pests.

[0061] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0062] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A deep learning-based intelligent inspection method for garden pests, characterized in that, Includes the following steps: Collect image data of vegetation surfaces within the garden inspection area; A pest identification model for the garden inspection area is constructed based on deep learning. Based on the pest identification model, pest identification is performed on the image data to obtain the pest attribute characteristics and pest distribution information on the vegetation in the garden inspection area. The environmental parameters within the garden inspection area are monitored, and the pest aggregation is divided into different areas based on the environmental parameters and the pest distribution information. Each aggregation area of ​​pests in the garden inspection area is obtained. The path association between each aggregation area is determined based on deep learning and the preset pest inspection path in the garden inspection area. Based on the pest attributes and the path relationships between various clusters, the preset pest inspection paths within the garden inspection area are adjusted to obtain differentiated inspection paths for each cluster. Pest inspections are conducted in the garden inspection area using differentiated inspection routes for each cluster area.

2. The method as described in claim 1, characterized in that, The specific steps of building a pest identification model for garden inspection areas based on deep learning include: Acquire historical pest image data within the garden inspection area; The historical pest image data is divided into training set, validation set and test set to construct a pest identification model in the garden inspection area based on deep learning.

3. The method as described in claim 1, characterized in that, Based on the pest identification model, pest identification is performed on the image data to obtain pest attribute characteristics and distribution information on vegetation within the garden inspection area, specifically including: The image data is loaded into the pest identification model, and pest identification is performed on each image in the image data to obtain pest information for each image. Based on all pest information, determine the pest characteristics and distribution information of the vegetation in the garden inspection area.

4. The method as described in claim 1, characterized in that, Based on the environmental parameters and the pest distribution information, the garden inspection area is divided into pest aggregation zones, resulting in the following specific pest aggregation zones within the garden inspection area: The environmental parameters are standardized and preprocessed to obtain the preprocessed environmental parameters; The criteria for classifying pest clusters in the garden inspection area are determined based on the environmental parameters and the pest distribution information. Based on the aforementioned criteria for classifying pest clusters, the garden inspection area is divided to obtain various clusters of pests within the garden inspection area.

5. The method as described in claim 1, characterized in that, Based on deep learning and pre-defined pest inspection paths within the garden inspection area, the path relationships between various clustering areas during the pest inspection process are determined, specifically including: Obtain the preset pest inspection path within the garden inspection area; Determine the intersection characteristics of the preset pest inspection path with each gathering area during the pest inspection process; Based on deep learning and the aforementioned intersection features, the path associations between various aggregation areas are determined during pest inspection.

6. The method as described in claim 1, characterized in that, Based on the pest attributes and path relationships between different clusters, the preset pest inspection paths within the garden inspection area are adjusted to differentiate them, resulting in differentiated inspection paths for each cluster. Specifically, these differentiated inspection paths include: Based on the characteristics of the pests, determine the inspection schedule information for each cluster area; Based on the path relationships between various cluster areas, determine the path differentiation adjustment information within the garden inspection area; By using the inspection arrangement information and the path differentiation adjustment information, the preset pest inspection paths within the garden inspection area are differentiated to obtain differentiated inspection paths for each cluster area.

7. The method as described in claim 1, characterized in that, Pest inspection in garden areas is conducted using differentiated inspection routes for different clusters of pests. This includes: Input the differentiated inspection paths of each cluster area into the inspection equipment of the garden inspection area; The inspection equipment performs pest inspections on the garden inspection area according to the differentiated inspection paths of each cluster area.

8. A deep learning-based intelligent inspection device for garden pests, characterized in that, include: The data acquisition module is used to collect image data of the vegetation surface within the garden inspection area; The processing module is used to construct a pest identification model in the garden inspection area based on deep learning, and to identify pests in the image data based on the pest identification model to obtain the pest attribute characteristics and pest distribution information on the vegetation in the garden inspection area. The processing module is also used to monitor environmental parameters within the garden inspection area, divide the garden inspection area into pest clusters based on the environmental parameters and the pest distribution information, obtain various clusters of pests within the garden inspection area, and determine the path association between various clusters during the pest inspection process based on deep learning and the preset pest inspection path within the garden inspection area. The processing module is also used to differentiate the preset pest inspection path in the garden inspection area based on the pest attribute characteristics and the path association between each cluster area, so as to obtain the differentiated inspection path for each cluster area. The execution module is used to conduct pest inspections in the garden inspection area through differentiated inspection paths for each cluster area.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the deep learning-based intelligent inspection method for garden pests as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent inspection method for garden pests based on deep learning as described in any one of claims 1 to 7.

Images