Tea disease and pest recognition system and device based on image recognition target detection technology

Abstract

The invention relates to the technical field of tea disease and insect pest recognition, in particular to a tea disease and insect pest recognition system and device based on an image recognition target detection technology. The tea plant disease and insect pest identification system comprises a path setting module used for screening out an optimal detection path passing through all photographing positions; and the target recognition module is used for constructing a deep convolutional neural network target recognition model based on parrot algorithm optimization and carrying out disease and pest recognition on the preprocessed image. The unmanned aerial vehicle is used for photographing, so that the target tea image can be quickly and comprehensively photographed, and high-resolution pest and disease damage image data can be obtained. Through an optimization method based on a He-Make algorithm, a globally optimal detection path can be found. Meanwhile, hyper-parameters of the deep convolutional neural network model are optimized through the parrot algorithm, the model performance is improved, the optimized deep convolutional neural network model can better learn the pest and disease damage characteristics of the tea image, and the recognition accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tea plant pest identification, and in particular to a tea plant pest identification system and device based on image recognition target detection technology. Background Art

[0002] Tea plants are susceptible to numerous pests and diseases, including powdery mildew, which manifests as a white powdery substance on the surface of tea leaves, severely impacting quality and yield. Anthracnose, manifests as brown, irregular spots on leaves, which in severe cases can cause them to wilt and fall off. Red spider mite, which develops tiny white spots on the leaf surface and can cause leaves to dry and curl. Gray spot, which develops grayish-brown spots on leaves and can spread widely, affecting the entire plant. Pests include: Tea aphids, which feed on leaf sap in large numbers, causing leaves to turn yellow, become deformed, and have their growth stunted. Tea geometrids, whose larvae feed on young leaves, creating irregular holes. Tea green leafhoppers, which feed on sap from leaves and branches, causing spots and deformities. Tea mites, which infest the underside of leaves in large numbers, can cause them to dry out and fall off. Tea armyworms, whose larvae feed on new shoots and leaves, severely impacting the plant's growth and development. These diseases directly impact tea quality and can lead to significant reductions in yield. Pests suck the sap of tea plants, stunting their growth and reducing yields. Severe pests and diseases can cause tea plants to wither and die. Pests and diseases can also cause secondary problems, such as affecting the growth of new shoots the following year. Therefore, timely detection and effective prevention and control of tea pests and diseases are crucial to ensuring tea quality and yield.

[0003] To address these pest and disease challenges, a tea pest and disease identification system is often needed to promptly detect and implement preventive measures, minimizing the impact of pests and diseases on tea production. However, due to the large size and complex terrain of tea plantations, accurate and comprehensive pest and disease identification and testing of tea leaves within the plantations is often impossible. Furthermore, conventional identification methods struggle to ensure accurate detection, impacting tea quality and yield. Summary of the Invention

[0004] Based on this, it is necessary to propose a tea pest and disease identification system and device based on image recognition target detection technology to address the problem that tea gardens are large in area, have relatively complex terrain, and are difficult to achieve accurate identification and detection.

[0005] The present invention is implemented by the following technical solution: a tea pest and disease identification system based on image recognition target detection technology comprises the following steps: The tea pest and disease identification system based on image recognition target detection technology is used to identify tea pests and diseases in tea gardens. The identification system includes: The path setting module includes a map construction unit, a photo location selection unit, and a path planning unit. The map construction unit is used to construct a three-dimensional map of the tea garden. The photo location selection unit is used to set multiple photo locations for the three-dimensional map. The path planning unit is used to construct an initial locally optimal motion detection path using a greedy algorithm. The greedy algorithm is optimized using the Hippo algorithm and a mutation operation is introduced to select the optimal motion detection path.

[0006] The data acquisition module includes an image acquisition unit and a position data acquisition unit. The image acquisition unit is used to capture the image of the target tea leaves in real time. The position data acquisition unit is used to synchronously record the coordinates of the target tea leaves.

[0007] The image preprocessing module is used to preprocess the target tea image to obtain a preprocessed image.

[0008] The target recognition module is used to build a deep convolutional neural network target recognition model optimized based on the Parrot algorithm to identify pests and diseases in preprocessed images.

[0009] The result feedback module includes an information labeling unit, a data upload unit, and a communication unit. The information labeling unit is used to label target tea images identified as having pests and diseases, generating corresponding labeled images. The data upload unit is used to upload the labeled images to a database. The communication unit is used to send the labeled images to tea garden managers.

[0010] By selecting detection and photographing locations and using drones for photography, the drones can quickly and comprehensively capture target tea leaves, acquiring high-resolution pest and disease image data. Compared to manual inspections, drones can cover a wider area, improving data collection efficiency. An optimization method based on the Hippo algorithm effectively addresses the problem of greedy algorithms easily falling into local optima and finds the globally optimal detection path. This significantly improves the scientific and targeted nature of tea garden pest and disease detection, providing more intelligent diagnostic services for tea farmers. Furthermore, the Parrot algorithm is used to optimize the hyperparameters of the deep convolutional neural network model, improving model performance. This optimized deep convolutional neural network model is able to better learn the pest and disease characteristics of tea images, improving recognition accuracy.

[0011] Furthermore, the specific method for the photo location selection unit to select multiple photo locations is: setting the number of photo locations according to the area of ​​the tea garden and the effective range that each photo can be taken by the drone, evenly distributing the photo locations in the tea garden, and obtaining the coordinates of each photo location.

[0012] Furthermore, the specific steps of the path planning unit using the greedy algorithm to generate the initial local optimal mobile detection path are as follows: The initial path is generated by the greedy algorithm as follows: Initialization: Set the starting and ending photo locations in the tea garden.

[0013] Select local optimum: Calculate the shortest distance from the starting location to all other undetected locations based on the coordinate information of the location, select the direction with the shortest distance, and move to the next location.

[0014] Update current photo location: Use the selected photo location as the starting photo location.

[0015] Output detection path: Repeatedly select the local optimal path to update the current photo position until the end photo position is reached.

[0016] Repeat the initialization to output the detection path, obtain multiple detection paths, and calculate the total detection time of the detection path.

[0017] Furthermore, the specific method of the path planning unit to optimize the greedy algorithm through the Hippo algorithm is as follows: Initialize the hippo population and randomly generate a group of hippo individuals, each of which represents a possible path improvement solution.

[0018] Define the fitness function, taking into account the path length and time factors, and its formula is:

[0019] in, 、 is the weight coefficient, L is the path length, and H is the time taken for the path.

[0020] The hippopotamus position is updated using the following formula:

[0021] in, and are the positions of the i-th hippopotamus at the t+1th and tth iterations in the j-th dimension, respectively. is the weight that controls the approach to the optimal individual, is the weight that controls random exploration. is the position of the best hippopotamus in the current iteration in the jth dimension, is the position of a hippopotamus randomly selected from the population in the jth dimension.

[0022] Furthermore, mutation operations: Mutate hippopotamus individuals with a certain probability. For the individuals to be mutated, mutation is achieved by adding a random small offset. The formula is:

[0023] in, Is a random offset value within a certain range.

[0024] From the updated and mutated hippo population, select the path represented by the individual with the best fitness.

[0025] It is compared and merged with the preliminary path generated by the greedy algorithm.

[0026] Repeat the Hippo algorithm optimization ~ select the optimal path until the preset maximum number of iterations is reached.

[0027] Determine and output the optimized optimal detection path.

[0028] Furthermore, the preprocessing process of the image preprocessing module is as follows: The image gradient is calculated using the Sobel operator to obtain the gradient magnitude for each pixel. The variance of the gradient magnitude is calculated. A determination is made as to whether the gradient magnitude variance is greater than a set variance threshold. If so, the corresponding target tea image is retained as the screening image. Image denoising, image enhancement, image normalization, geometric transformation, and image segmentation are performed on the screening image to obtain a preprocessed image.

[0029] Furthermore, the specific method of constructing the deep convolutional neural network detection model based on the Parrot algorithm optimization in the target recognition module is as follows: Obtain a tea image dataset and divide it into a training set, a validation set, and a test set according to a preset ratio.

[0030] Construct an initial pest and disease recognition model based on deep convolutional neural network.

[0031] The Parrot algorithm is used to optimize the hyperparameters of the initial pest and disease identification model to obtain the tea pest and disease identification model: The training set was used to train the deep convolutional neural network model, the validation set was used to evaluate the performance of the tea disease and pest recognition model, and the model parameters that met the recognition accuracy were retained. The test set was used to test the trained tea disease and pest recognition model, and the optimal model parameters were retained.

[0032] Furthermore, the target recognition module uses the Parrot algorithm to optimize the hyperparameters of the initial pest and disease recognition model. The specific steps are as follows: The hyperparameters of the deep convolutional neural network are mapped to the search space of the Parrot algorithm, where each Parrot corresponds to a set of hyperparameter combinations.

[0033] Initialize the positions of N parrots and the best position , and the optimal position of the group , use random initialization to set the initial value.

[0034] Update each parrot's position based on observed learning behavior , the formula is:

[0035] in, 、 is a random vector between [0, 1] that controls the strength of observational learning. is the position of each parrot at time t. is the optimal position of each parrot at time t. is the optimal position of the group at time t.

[0036] Update the optimal position of each parrot based on the memory learning behavior , the formula is:

[0037] in, () is the fitness function, which represents the loss of the selected hyperparameter combination on the validation set.

[0038] Update the optimal position of the group based on social learning behavior , the formula is:

[0039] Repeatedly map the hyperparameters of the deep convolutional neural network to the search space of the Parrot algorithm until the preset maximum number of iterations is reached.

[0040] The present invention also provides a device for identifying tea pests and diseases based on image recognition target detection technology, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The functional modules in the device are deployed using the aforementioned system for identifying tea pests and diseases based on image recognition target detection technology. When the processor executes the computer program, the aforementioned system for identifying tea pests and diseases based on image recognition target detection technology is implemented, thereby enabling pest and disease identification of target tea leaves within a tea garden, thereby improving the accuracy and reliability of pest and disease identification.

[0041] Compared with the prior art, the present invention has the following beneficial effects: The present invention selects a detection and photo-taking location and uses a drone to take photos. The drone can quickly and comprehensively capture a panoramic view of the tea garden and obtain high-resolution image data of pests and diseases. Compared with manual inspections, drones can cover a wider area, improving data collection efficiency. The optimization method based on the Hippo algorithm can effectively solve the problem that the greedy algorithm easily falls into the local optimum and find the globally optimal detection path. This greatly improves the scientific nature and pertinence of tea garden pest and disease detection and provides tea farmers with more intelligent diagnostic services. At the same time, the Parrot algorithm is used to optimize the hyperparameters of the deep convolutional neural network model, improve model performance, and enable the optimized convolutional neural network model to better learn the pest and disease characteristics of tea images and improve recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a framework diagram of the tea pest and disease identification system based on image recognition target detection technology in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component. When a component is considered to be "fixed to" another component, it may be directly fixed to the other component or there may be a central component.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0046] Example 1: Please refer to Figure 1 This embodiment provides a tea pest and disease identification system based on image recognition target detection technology, which is used to identify pests and diseases in tea images in a tea garden obtained by a drone, including: a path setting module, a data acquisition module, an image preprocessing module, a target recognition module and a result feedback module.

[0047] The path setting module includes a map construction unit, a photo location selection unit and a path planning unit. Among them, the map construction unit is used to construct a three-dimensional map of the tea garden; the photo location selection unit is used to set the locations where photos need to be taken in the tea garden; the path planning unit optimizes the greedy algorithm through the Hippo algorithm and introduces a mutation operation to obtain the optimal detection path passing through all photo locations.

[0048] In this embodiment, the map construction unit is used to accurately map the tea garden using GPS or manual measurement to construct a three-dimensional map. Based on the garden's area and the effective range of each drone photo, a number of photo locations is determined. These locations are evenly distributed throughout the garden, and the coordinates of each location are determined based on the three-dimensional map. The selection of photo locations ensures comprehensive and accurate detection of pests and diseases within the garden, providing a strong basis for timely prevention and control measures.

[0049] The path planning unit is used to take each location where a photo is required as a necessary point on the detection path, obtain the detection route with the shortest distance passing through each photo location through a greedy algorithm, and calculate the total detection time for the drone to pass through the detection route.

[0050] Among them, the method of obtaining the shortest detection path through the greedy algorithm is: Initialization: Set the starting and ending photo locations in the tea garden; Select local optimum: Calculate the shortest distance from the starting location to all other undetected locations based on the coordinate information of the location, select the direction with the shortest distance, and move to the next location. Update the current photo location: use the selected photo location as the starting photo location; Output detection path: Repeatedly select the local optimum to update the current photo position until the end photo position is reached; Repeat the initialization to output the detection path, obtain multiple detection paths, and calculate the total detection time of the detection path.

[0051] In this embodiment, the distance and time of the detection path are used as reference factors for the detection path, and the Hippo algorithm greedy algorithm is used and a mutation operation is introduced. The specific method is as follows: Hippo algorithm optimization, including: Initialize the hippo population and randomly generate a group of hippo individuals, each of which represents a possible path improvement solution; Define the fitness function, taking into account the path length and time factors, and its formula is:

[0052] in, 、 is the weight coefficient, L is the path length, and H is the time taken for the path; Hippo Location Update: The update formula is:

[0053] in, and are the positions of the i-th hippopotamus individual in the j-th dimension at the t+1th and tth iterations respectively; is the weight that controls the approach to the optimal individual, is the weight that controls random exploration; is the position of the best hippopotamus in the current iteration in the jth dimension, is the position of a hippopotamus randomly selected from the population in the jth dimension; Mutation operation: mutate the hippopotamus individuals with a certain probability. For the individuals to be mutated, mutation is achieved by adding a random small offset. The formula is:

[0054] in, Is a random offset value within a certain range; Select the optimal path, including: From the updated and mutated hippo population, select the path represented by the individual with the best fitness; Compare and merge it with the preliminary path generated by the greedy algorithm; Repeat the optimization process: Repeat the Hippo algorithm optimization ~ select the optimal path until the preset maximum number of iterations is reached.

[0055] Output the final detection path: Determine and output the optimized shooting path for actual tea pest and disease detection.

[0056] The data acquisition module includes an image acquisition unit and a position data acquisition unit. The image acquisition unit is used to capture the target tea leaf image in real time; the position data acquisition unit is used to synchronously record the target tea leaf coordinates.

[0057] In this embodiment, a high-resolution camera can be selected to clearly capture subtle signs of pests and diseases and damage on tea leaves. A certain optical zoom function is also required to facilitate magnified observation of specific areas without being close to the tea garden, allowing for more accurate detection of pests and diseases. Tea gardens may have uneven lighting, so the selected camera must have good low-light performance to ensure clear images even on cloudy days or in shaded areas. Furthermore, accurate color reproduction is required to help distinguish subtle color differences between normal tea leaves and tea leaves affected by pests and diseases.

[0058] The image preprocessing module is used to preprocess the target tea image to obtain a preprocessed image.

[0059] In this embodiment, the preprocessing process of the image preprocessing module is specifically as follows: The gradient of the image is calculated using the Sobel operator to obtain the gradient amplitude of each pixel; the gradient amplitude variance is calculated; and it is determined whether the gradient amplitude variance is greater than a set variance threshold. If so, the corresponding target tea image is retained as the screening image.

[0060] Through the above process, blurred and unclear images can be removed, ensuring that the images used in subsequent processing and analysis have good visual effects and information accuracy.

[0061] The screened image is subjected to image denoising, image enhancement, image normalization, geometric transformation, and image segmentation to obtain a preprocessed image. Image denoising is used to reduce noise interference in the image, making the image details and edges clearer and improving the accuracy of subsequent processing.

[0062] Image enhancement is used to enhance the contrast and brightness of an image, making the features in the image more obvious and facilitating subsequent analysis and recognition.

[0063] Image normalization is used to standardize the pixel values ​​of an image so that their mean and variance fall within a specific range. This eliminates differences in brightness and contrast between images, making subsequent processing more stable and reliable.

[0064] Geometric transformation corrects the posture and position of images through rotation and scaling, making it easier to compare and process different images.

[0065] Image segmentation is used to divide an image into different regions so that different regions can be processed and analyzed separately.

[0066] Obtaining a preprocessed image through the above-mentioned image preprocessing method can improve the quality and usability of the image, provide a better foundation for subsequent image recognition tasks, and thus improve the performance and accuracy of the entire image recognition system.

[0067] The target recognition module is used to build a deep convolutional neural network target recognition model optimized based on the Parrot algorithm, and use this model to identify pests and diseases in preprocessed images.

[0068] In this embodiment, the specific manner in which the target recognition module constructs a deep convolutional neural network detection model optimized based on the Parrot algorithm is as follows: Collect a large dataset of tea images, including images of healthy tea leaves and tea leaves infected with pests and diseases, covering different tea varieties and pest types; annotate the images to indicate the presence of pests and diseases and the type of pests and diseases; scale, crop, and flip the images to improve the generalization ability of the model; and divide the image dataset into training, validation, and test sets. An initial pest and disease recognition model based on a deep convolutional neural network is constructed, with a reasonable design of the input layer, convolution layer, pooling layer, and fully connected layer. The input layer is used to input the preprocessed image; the convolution layer is used to extract features from the input image, using the convolution kernel to slide on the input image, perform convolution operations, and generate feature maps; the pooling layer reduces the size of the feature map by performing downsampling operations on the feature map; the fully connected layer combines and fuses the local features extracted by the convolution layer to generate a high-level global feature representation, flattens the feature map into a one-dimensional vector, and calculates the output through matrix multiplication; the final fully connected layer uses the binary classification output to detect whether the target tea has pests and diseases, and if so, identifies the type of pests and diseases; The Parrot algorithm is used to optimize the hyperparameters of the initial pest and disease identification model to obtain a tea pest and disease identification model. The specific process is as follows: Map the hyperparameters of the deep convolutional neural network to the search space of the Parrot algorithm, where each Parrot corresponds to a set of hyperparameter combinations; Initialize the positions of N parrots and the best position , and the optimal position of the group , use random initialization to set the initial value; Update each parrot's position based on observed learning behavior , the formula is:

[0069] in, 、 is a random vector between [0, 1] that controls the intensity of observational learning; is the position of each parrot at time t; is the optimal position of each parrot at time t; is the optimal position of the group at time t; Update the optimal position of each parrot based on the memory learning behavior , the formula is:

[0070] in, () is the fitness function, which represents the loss of the selected hyperparameter combination on the validation set; Update the optimal position of the group based on social learning behavior , the formula is:

[0071] Repeatedly map the hyperparameters of the deep convolutional neural network to the search space of the Parrot algorithm until the preset maximum number of iterations is reached.

[0072] Use the optimized hyperparameters to train the deep convolutional neural network model on the training set; The cross entropy loss function is used to update the model parameters through the back propagation algorithm. The specific process is as follows: Forward propagation: input the input data into the deep convolutional neural network, and extract and calculate features through a series of convolutional layers, pooling layers, and fully connected layers to obtain the output prediction value of the model; The cross entropy loss function is used to calculate the loss between the model's predicted value and the true label. The formula of the cross entropy loss function is usually:

[0073] in, is the true label, is the predicted probability of the model; Starting from the output layer, calculate the error gradient of each neuron, and use the chain rule to backpropagate the error gradient from the output layer to the input layer layer by layer to calculate the gradient of each neuron; Based on the calculated gradient, the optimization algorithm is used to update the model parameters; The above steps are repeated continuously, using new training data for forward propagation, loss calculation, backpropagation and parameter update until the model reaches the predetermined number of training rounds.

[0074] During training, the model performance is regularly evaluated on the validation set, and early stopping is used to avoid overfitting. Evaluate the trained model on the test set and calculate the accuracy, recall and F1 value; Analyze the model's performance on different types of tea images and identify any existing problems; Based on the evaluation results, the model structure is further optimized or the hyperparameters are adjusted to improve the detection performance.

[0075] The trained model is applied to the tea pest and disease identification system, and the target tea image obtained by the data acquisition module is subjected to feature extraction to detect whether the target tea has pests and diseases, and if so, the type of pests and diseases is identified.

[0076] By constructing an object recognition module based on a deep convolutional neural network detection model optimized using the Parrot algorithm, and optimizing the model's parameters and structure, the model can more accurately capture the characteristics of tea pests and diseases, thereby reducing misidentifications and missed detections and providing accurate evidence for timely prevention and control measures. This allows the model to adapt to tea samples of varying varieties, growing environments, and pest and disease manifestations, maintaining strong recognition accuracy even when faced with new and previously unseen tea pest and disease conditions. The model can process large amounts of tea image data in a short period of time, enabling rapid pest and disease detection and improving work efficiency. This helps identify problems promptly and implement countermeasures, thereby reducing losses caused by pests and diseases. It facilitates accurate identification of pests and diseases in their early stages, enabling early implementation of preventive and control measures to prevent their spread and worsening. By replacing some manual inspection work, the model reduces reliance on professional expertise and expertise, saving manpower. Providing scientific data support to tea growers and relevant departments will help formulate more rational and targeted pest and disease control strategies, enabling precise application of pesticides, reducing pesticide use, and protecting the environment. Timely and effective identification and treatment of pests and diseases helps ensure the growth and development of tea, thereby improving tea quality and yield, and increasing economic benefits. Over the long term, the accumulation of a large amount of pest and disease identification data will provide a foundation for further research on the occurrence patterns and development trends of tea pests and diseases, and help promote the continuous advancement of tea pest and disease control technology.

[0077] The result feedback module includes an information labeling unit, a data uploading unit and a communication unit. The information labeling unit is used to mark the pest and disease type and location information of the target tea leaves identified with pests and diseases, and obtain the corresponding labeled image; the data uploading unit is used to upload the labeled image to the database; and the communication unit is used to send the labeled image to the tea garden managers.

[0078] In this embodiment, the annotated images are sent to tea garden managers via a communication unit, providing accurate information for subsequent pest control efforts. Uploading annotated images to the database increases its comprehensiveness, incorporating more pest and disease images, thereby broadening the model's recognition range and improving its accuracy. Rich and diverse data allows the model to learn more features and patterns, thereby improving the accuracy and robustness of its recognition and detection. Furthermore, as new image features, object types, or application scenarios emerge over time, supplementing the database allows the model to keep pace with these changes, maintaining its effectiveness and practicality.

[0079] In order to implement the aforementioned tea pest and disease identification system based on image recognition target detection technology and apply it to existing tea pest and disease identification processes, thereby achieving accurate and reliable pest and disease identification for target tea images within a tea garden, this embodiment further provides a tea pest and disease identification device based on image recognition target detection technology. The tea pest and disease identification device includes a memory, a processor, and a computer program stored in the memory and executable by the processor. When the processor executes the computer program, it implements the aforementioned tea pest and disease identification system based on image recognition target detection technology, thereby achieving pest and disease identification for target tea images within a tea garden, thereby improving the accuracy and reliability of pest and disease identification.

[0080] The computer device may be a smartphone, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server, or cabinet server (including a standalone server or a server cluster consisting of multiple servers) that can execute programs. The computer device of this embodiment includes at least, but is not limited to, a memory and a processor that can be interconnected via a system bus.

[0081] In this embodiment, the memory (i.e., readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of a computer device, such as the hard disk or internal memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk equipped with the computer device, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Of course, the memory may also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory is generally used to store the operating system and various application software installed on the computer device. In addition, the memory may also be used to temporarily store various types of data that have been output or are about to be output.

[0082] In some embodiments, the processor can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. Such a processor is typically used to control the overall operation of a computer device. In this embodiment, the processor is used to execute program code stored in a memory or process data, thereby implementing the aforementioned tea pest and disease identification system based on image recognition target detection technology, thereby achieving accurate and reliable pest and disease identification of target tea images within a tea garden.

[0083] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A tea pest and disease identification system based on image recognition target detection technology is used to identify tea pests and diseases in tea gardens, and is characterized by: include: A path setting module, comprising a map construction unit, a photo location selection unit and a path planning unit, wherein the map construction unit is used to construct a three-dimensional map of the tea garden; The photographing position selection unit is used to set a plurality of photographing positions of the three-dimensional map; the path planning unit is used to generate an initial local optimal mobile detection path by using a greedy algorithm, optimize the greedy algorithm by using a hippopotamus algorithm and introduce a mutation operation, and screen out an optimal mobile detection path; A data acquisition module, including an image acquisition unit and a position data acquisition unit, wherein the image acquisition unit is used to capture the target tea leaf image in real time; The position data acquisition unit is used to synchronously record the coordinates of the target tea leaves; An image preprocessing module, used for preprocessing the target tea image to obtain a preprocessed image; A target recognition module, used to construct a deep convolutional neural network target recognition model based on the Parrot algorithm optimization, and to perform pest and disease recognition on the pre-processed image; The result feedback module includes an information marking unit, a data uploading unit and a communication unit. The information marking unit is used to mark the target tea images identified with pests and diseases to obtain corresponding marked images; the data uploading unit is used to upload the marked images to a database; and the communication unit is used to send the marked images to tea garden managers.

2. The tea pest and disease identification system based on image recognition target detection technology according to claim 1 is characterized in that: The specific method in which the photographing position selection unit selects multiple photographing positions is as follows: the number of photographing positions is set according to the area of ​​the tea garden and the effective range that each photo can be taken by the drone, the photographing positions are evenly distributed in the tea garden, and the coordinates of each photographing position are obtained.

3. The tea pest and disease identification system based on image recognition target detection technology according to claim 1 is characterized in that: The specific steps of the path planning unit using the greedy algorithm to generate the initial local optimal mobile detection path are: Initialization: Set the starting and ending photo taking positions in the tea garden; Select local optimum: Calculate the shortest distance from the starting point to all other undetected locations based on the coordinate information of the location, select the direction with the shortest distance, and move to the next location; Update the current photo location: use the selected photo location as the starting photo location; Output detection path: Repeatedly select the local optimum to update the current photo position until the end photo position is reached; Repeat the initialization to output the detection path, obtain multiple detection paths, and calculate the total detection time of the detection path.

4. The tea pest and disease identification system based on image recognition target detection technology according to claim 3 is characterized in that: The specific method of the path planning unit optimizing the greedy algorithm through the Hippo algorithm is: Initialize the hippo population and randomly generate a group of hippo individuals, each of which represents a possible path improvement solution; Define the fitness function, taking into account the path length and time factors, and its formula is: in, , is the weight coefficient, L is the path length, and H is the time taken for the path; Hippo position update, the update formula is: in, and are the positions of the i-th hippopotamus individual in the j-th dimension at the t+1th and tth iterations respectively; is the weight that controls the approach to the optimal individual, is the weight controlling random exploration; is the position of the best hippopotamus in the current iteration in the jth dimension, is the position of a hippopotamus randomly selected from the population in the jth dimension.

5. The tea pest and disease identification system based on image recognition target detection technology according to claim 3 is characterized in that: Mutation operation: The hippopotamus individuals are mutated with a certain probability. For the individuals to be mutated, a small random offset is added to achieve mutation. The formula is: in, is a random offset value within a certain range; From the updated and mutated hippo population, select the path represented by the individual with the best fitness; Compare and merge it with the preliminary path generated by the greedy algorithm; Repeat the Hippo algorithm optimization and select the optimal path until the preset maximum number of iterations is reached; Determine and output the optimized optimal detection path.

6. The tea pest and disease identification system based on image recognition target detection technology according to claim 1 is characterized in that: The preprocessing process of the image preprocessing module is as follows: The gradient of the image is calculated by the Sobel operator to obtain the gradient amplitude of each pixel; the gradient amplitude variance is calculated; it is determined whether the gradient amplitude variance is greater than a set variance threshold, and if so, the corresponding target tea image is retained as a screening image; image denoising, image enhancement, image normalization, geometric transformation and image segmentation are performed on the screening image to obtain a preprocessed image.

7. The tea pest and disease identification system based on image recognition target detection technology according to claim 1 is characterized in that: The specific method of constructing the deep convolutional neural network detection model based on the Parrot algorithm optimization in the target recognition module is as follows: Obtaining a tea image dataset, and dividing the tea image dataset into a training set, a validation set, and a test set according to a preset ratio; Construct an initial pest and disease identification model based on deep convolutional neural network; The Parrot algorithm is used to optimize the hyperparameters of the initial pest and disease identification model to obtain a tea pest and disease identification model; The training set was used to train the deep convolutional neural network model, the validation set was used to evaluate the performance of the tea disease and pest recognition model, and the model parameters that met the recognition accuracy were retained. The test set was used to test the trained tea disease and pest recognition model, and the optimal model parameters were retained.

8. The tea pest and disease identification system based on image recognition target detection technology according to claim 7 is characterized in that: The specific steps of optimizing the hyperparameters of the initial pest identification model by the target identification module using the parrot algorithm are as follows: Map the hyperparameters of the deep convolutional neural network to the search space of the parrot algorithm, with each parrot corresponding to a set of hyperparameter combinations; Initialize the positions of N parrots and the best location , and the optimal position of the group , use random initialization to set the initial value; Update the position of each parrot based on observed learning behavior , the formula is: in, , is a random vector between [0, 1], controlling the intensity of observational learning; is the position of each parrot at time t; is the optimal position of each parrot at time t; is the optimal position of the group at time t; Update the optimal position for each parrot based on the memory learning behavior , the formula is: in, () is the fitness function, which represents the loss of the selected hyperparameter combination on the validation set; Update the optimal position of the group based on social learning behavior , the formula is: Repeatedly map the hyperparameters of the deep convolutional neural network to the search space of the Parrot algorithm until the preset maximum number of iterations is reached.

9. A tea pest and disease identification device based on image recognition target detection technology, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: Each functional module in the tea pest and disease identification device is deployed in the form of a tea pest and disease identification system based on image recognition target detection technology as described in any one of claims 1 to 8, so as to realize pest and disease identification of target tea images in a tea garden.

Images