Intelligent field rice planthopper identification method and system based on AR binocular glasses

Abstract

The invention relates to the technical field of agricultural disease and insect pest monitoring, in particular to a field rice planthopper intelligent identification method and system based on AR binocular glasses. According to the method, a white disc image of rice planthoppers is collected, the image is cut into small-area sub-images before identification, and the proportion of low-age nymphs is increased; calculating key data such as the amount of hectoria; the model training process focuses on learning features of low-age nymphs; according to the system, an identification model, a calculation module and a storage module are integrated and installed in local hardware of AR glasses, so that the limitation of a network is eliminated; and the identification result is superposed on a display screen of the glasses in real time. According to the method, the detection rate and the recognition accuracy of millimeter-level low-age nymphs are improved; according to the invention, millisecond-level rice planthopper identification and counting in a field are realized, the problem of network dependence is overcome, and an identification result and a real scene are superposed in real time through AR display, so that investigators can obtain what you see.

Description

Technical Field

[0001] This invention relates to the field of agricultural pest and disease monitoring technology, specifically to a method and system for intelligent identification of rice planthoppers in the field based on AR binocular glasses. Background Technology

[0002] Rice planthoppers are major migratory pests in rice production, characterized by their tiny size (0.1-4 mm), rapid reproduction, and explosive outbreaks. They can lead to reduced rice yields or even crop failure, posing a serious threat to food security. Accurate monitoring and forecasting are prerequisites for precise and green pest control. Current monitoring methods mainly rely on the "Rice Planthopper Monitoring and Survey Specification GB / T 15794-2009," employing methods such as white disc tapping and manual identification and counting. This method suffers from several drawbacks: low efficiency, limited manual survey scope, difficulty in achieving large-scale and rapid monitoring; high skill requirements for personnel, especially in identifying different instars of rice planthoppers, which requires extensive experience and is susceptible to subjective factors leading to misjudgments; poor real-time performance, as manual survey data requires subsequent processing and analysis, failing to provide immediate guidance for pest and disease control; and high labor intensity, with complex field environments leading to fatigue among personnel working long hours.

[0003] With the development of IoT, AI, and AR technologies, the application of AR and AI image recognition technologies in agriculture has gradually deepened. Several solutions combining image recognition technologies have emerged. AR glasses can overlay virtual information onto the real world, achieving a "what you see is what you get" interactive experience. Target detection models, due to their speed and high accuracy, have been widely used in real-time object recognition. However, existing recognition models are mostly designed for larger adult insects, while rice planthoppers are tiny (nymphs 0.1-3mm, adults 3-4mm), and the morphological differences between different age groups are subtle. Directly applying existing models to field identification can easily lead to low recognition accuracy due to insufficient target pixels and inadequate extraction of age-related features. Furthermore, most intelligent recognition solutions rely on cloud processing, and network anomalies can easily lead to data loss, making it difficult to meet the needs of real-time, efficient, and high-precision monitoring of rice planthoppers in complex field environments. Summary of the Invention

[0004] The purpose of this invention is to provide a smart identification system and method for rice planthoppers in the field based on AR binocular glasses, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for intelligent identification of rice planthoppers in the field based on AR binocular glasses includes the following steps: S1. White Plate Image Acquisition: The investigators wear the AR binocular glasses and use a white plate to tap the rice clumps in the field. The acquisition is started by voice command or manual touch control module. The white plate image of the rice planthopper is acquired through the camera to ensure that the image contains a clear rice planthopper target. S2. Image preprocessing: The image block processing module performs noise reduction and grayscale normalization on the white disk image, and cuts it into multiple overlapping small sub-images with the same size as the model training stage, with an overlap rate of 30%~50%. S3. Target Recognition and Counting: The trained convolutional neural network model is integrated into the local processor of the AR binocular glasses to recognize the sub-images. The recognition process includes insect species, wing type, quantity, and age identification, specifically: Determine whether the target is a rice planthopper, and then identify the rice planthoppers step by step. To determine the species of the target insect, the insect species is determined based on body color, markings and wing vein patterns, while the wing type is determined based on the ratio of wing length to body length. Among them, a wing length / body length ratio ≥ 0.8 is judged as a long-winged adult, and a wing length / body length ratio < 0.5 is judged as a short-winged adult; Determine the age of the target insect by identifying whether it is an adult or a nymph based on the relative size of its body outline and the completeness of its wing development; If the target is an adult insect, label it with a category tag that combines the insect species and wing type; If the target is a nymph, its age can be initially determined based on its body shape and absolute size. Among them, a body length of 2-3mm is considered advanced age, and a body length of 0.1-2mm is considered young age; If it is an older nymph, label it with a category tag that combines the insect species and wing type; If it is a young nymph, a comprehensive judgment is made based on the precise size and subtle body color differences provided by the white disc image, and the category label is marked after eliminating the interference of impurities. After recognition is completed, record the bounding box coordinates and category label of each target in its respective sub-image; The category labels include long-winged adult white-backed planthopper (M-WBPH), short-winged adult white-backed planthopper (B-WBPH), older nymph white-backed planthopper (S-WBPH), long-winged adult brown planthopper (M-BPH), short-winged adult brown planthopper (B-BPH), older nymph brown planthopper (S-BPH), long-winged adult gray planthopper (M-SBPH), short-winged adult gray planthopper (B-SBPH), older nymph gray planthopper (S-SBPH), and young nymphs (J-RPH).

[0006] The recognition results of the sub-images are mapped back to the global coordinate system of the original image through the image coordinate mapping algorithm. The deduplication rule with an intersection-union ratio threshold of ≥0.3 is used to merge targets with overlapping positions and the same analog label in the global coordinate system, so as to obtain the recognition results of each rice planthopper individual in the whole image and complete the automatic counting of each category and the calculation of the proportion of each insect age. S4. Calculation of insect count per 100 clumps: The insect count per 100 clumps calculation module automatically calculates the insect count per 100 clumps according to the formula: insect count per 100 clumps = (total number of insects identified / number of clumps investigated) × 100, based on the total number of insects identified in step S3 and the number of clumps investigated. The number of clumps investigated is estimated through image analysis or input by the user. S5. Data storage and real-time display: The data storage and management module writes all the data from this survey into the local non-volatile memory integrated in the AR binocular glasses for storage and automatically generates a survey report; The insect species and quantity in the identification results, as well as the number of insects per hundred clusters, are superimposed in real time on the display screen of the AR binocular glasses through the binocular near-eye display module, and superimposed on the real white disk image seen by the investigator in the form of a semi-transparent floating window; Among them, the coordinate deviation between the superimposed position and the target coordinates of the rice planthopper in the real white disk image is ≤5 pixels; S6. Uploading and Visualizing Survey Results: After the survey is completed or when network access is available, the communication module will upload the survey report and all data in batches to the cloud platform; the survey report can be viewed via a mobile app or a web browser on a computer.

[0007] Furthermore, the training method for the convolutional neural network model includes the following steps: The first step is dataset construction and annotation. White disc images of multiple species of rice planthoppers in multiple scenes are collected using the AR binocular glasses. Annotation software is used to draw bounding boxes based on the morphological features of each rice planthopper in the images, and a corresponding category label is set for each bounding box. The annotation software includes LabelImg, with an annotation accuracy of ≤2 pixels for bounding box position error;

[0008] The second step is dataset preprocessing, which cuts the labeled white disk image into multiple overlapping small region sub-images with an overlap rate of 30% to 50%, in order to increase the number of training samples and force the model to focus on learning local subtle features. Among them, the original white disk image resolution is ≥8000×6000, the size of the small region sub-image is 512×512 pixels, cutting it into sub-images can increase the proportion of young nymph targets in the sub-image by 2-3 times, effectively increasing the number of training samples to 4-6 times that of the original samples. The third step is model training and validation. The YOLOv11 model is trained using the dataset from the second step, enabling the model to learn end-to-end. It jointly learns the color, texture, shape contour of a single frame image and the hidden depth information and relative size information of the target in the white disk image. The texture structure includes the density of punctures on the adult wing surface and the distribution of setae on the nymph body surface. The model was optimized using stochastic gradient descent with 300 iterations and an initial learning rate of 0.01, which decayed by a factor of 10 every 100 iterations. Training was stopped when the loss function value on the validation set did not decrease for 10 consecutive iterations. Among them, the loss function uses CIoU loss superimposed with Focal loss to solve the problem of imbalance between bounding box regression and young nymph samples; The fourth step is model evaluation and optimization: the trained model is evaluated using a test set, with segmentation accuracy, recall, and mAP as the core evaluation metrics; for young nymphs, a hard sample mining strategy is used for secondary fine-tuning until the recognition accuracy of each nymph age meets the preset requirements. Among them, the young nymphs are small in size and have indistinct morphological features, which can easily lead to low recognition accuracy. Therefore, a hard sample mining strategy is adopted during training to increase the training weight of young nymph samples. The final model achieved an accuracy of ≥95% for young nymphs, ≥97% for older nymphs, and ≥98% for adults on the test set.

[0009] The present invention also provides a field rice planthopper intelligent identification system based on AR binocular glasses, for implementing the aforementioned method, the system comprising: The image acquisition module is used to acquire a white disk image based on the AR binocular glasses, wherein the white disk image is an RGB image; The local identification and calculation module includes a trained convolutional neural network model and a module for calculating the number of insects per 100 clusters. The convolutional neural network model is trained to identify insect species, quantity, wing type, and age. The module for calculating the number of insects per 100 clusters is used to automatically calculate the number of insects per 100 clusters and the proportion of each age. The data storage and management module is used to automatically classify the survey data and securely write it into local non-volatile memory in a structured format for storage, and automatically generate a survey report based on the data; the data includes original images, processed images, recognition results, number of insects per hundred clusters, geographical location, survey timestamp, and surveyor ID; The data synchronization and display module includes a communication module, a cloud platform, and a display terminal. The communication module is used to upload locally stored data to the cloud platform. The cloud platform is used to receive, store, and manage the data. The display terminal includes a mini-program and a web terminal for visually displaying and sharing survey reports.

[0010] Furthermore, the binocular vision acquisition module includes a camera, a binocular near-eye display module, a positioning module, and a control module; the camera is installed on the upper front of the AR binocular glasses and above the binocular near-eye display module, and is used to acquire RGB images of the white disc of the rice planthopper; the binocular near-eye display module is used to overlay and display the recognition results of the convolutional neural network model; the positioning module is used to obtain the geographical location information of the survey location; and the control module is used to realize the interactive operation between personnel and the system, and the control module includes a voice recognition module and touch buttons.

[0011] Preferably, the camera has a resolution of 4800w, a focal length of 5mm, and supports autofocus; the binocular near-eye display module has a resolution of 2560×1440 and a field of view of 80°; the positioning module includes a GPS positioning module with a positioning accuracy of ≤10m.

[0012] Furthermore, the local recognition and calculation module also includes an image segmentation processing module and a coordinate mapping and deduplication algorithm module; the image segmentation processing module is used to adjust the white disk image and cut it into multiple overlapping small sub-images of the same size as the model training stage, with an overlap rate of 30%~50%; the coordinate mapping and deduplication algorithm module is used to map the recognition result back to the global coordinate system of the original image and delete duplicate recognized targets.

[0013] Preferably, the local non-volatile memory is a 1TB UFS 3.1; Furthermore, the convolutional neural network model and model file are stored in the local non-volatile memory; Furthermore, the insect count calculation module per hundred clumps supports manual input of the number of clumps to be surveyed or automatic retrieval of the number of rice clumps in the identification results to determine the number of clumps to be surveyed; Furthermore, the data storage and management module writes data to local non-volatile memory according to the directory structure of "date-plot-survey point", and adopts a log structure file system to ensure the atomicity of data writing. In the event of an unexpected power outage, the data that has been written can be retained. Furthermore, the communication module supports 4G, 5G, Wi-Fi and Bluetooth, and uses a breakpoint resume mechanism to upload data to the cloud platform; Furthermore, the various modules of the system are integrated and installed in the AR binocular glasses hardware body, which also contains a processor and a power module.

[0014] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows: 1. High recognition accuracy: Before training, the white disk image is cut into small sub-images, which can enhance the YOLOv11 model's recognition training for young nymphs with fewer pixels and force the model to pay more attention to learning difficult samples. This improves the system's detection rate of 0.1mm-level young nymphs to over 95%, and the average recognition accuracy of 10 types of rice planthopper targets is ≥97%, raising the detection rate and recognition accuracy of millimeter-level and even sub-millimeter-level young nymphs to a practical level.

[0015] 2. High survey efficiency: Utilizing the powerful local computing capabilities of AR glasses, millisecond-level identification and counting of rice planthoppers in the field are achieved, overcoming the network dependency problem. The entire identification process takes ≤3 seconds per plate, which is 5-8 times more efficient than manual counting.

[0016] 3. Intelligent data management: It realizes full-process automation of survey data, from image acquisition, recognition, counting to insect count per 100 plants and AR display, local storage and cloud synchronization, avoiding errors from manual recording and ensuring the integrity, security and traceability of data; it supports multi-dimensional data analysis and visualization, providing data support for early warning and prevention of rice planthoppers.

[0017] 4. Excellent real-time interactivity: By overlaying the recognition results with the real scene through AR display, investigators can obtain key information such as target location and quantity intuitively, which improves the work experience and decision-making speed.

[0018] 5. Technological Integration and Innovation: Creatively integrates AR vision with professional agricultural pest monitoring and forecasting, enhancing the model's ability to identify small targets and effectively improving its applicability for field surveys of rice planthoppers. Attached Figure Description

[0019] Figure 1 This is a flowchart of the identification method of the present invention; Figure 2 This is a schematic diagram of the identification system of the present invention; Figure 3 Data storage flowchart; Figure 4 This is a display effect diagram showing the real-time overlay effect; Figure 5 Show the preview of the mini-program; Figure 6 This is a screenshot showing the effect on a computer web interface. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to embodiments: like Figure 1 A method for intelligent identification of rice planthoppers in the field based on AR binocular glasses includes the following steps: S1. White Plate Image Acquisition: Researchers wear AR binocular glasses and select survey points in the field using the "five-point sampling method". They use a white plate to tap the rice clumps and start the acquisition by voice command "take a picture" or manual touch control button. The white plate image of rice planthopper is acquired through a 4800w resolution camera at a frame rate of 18fps. The image format is JPEG to ensure that the image contains a clear rice planthopper target. S2. Image preprocessing: The image block processing module first performs noise reduction and grayscale normalization on the white disk image, normalizes the pixel value to 0.5, and then cuts it into multiple overlapping small sub-images with a size of 512×512 pixels and an overlap rate of 30%. S3. Target Recognition and Counting: The trained convolutional neural network model YOLOv11 is used to recognize sub-images. The convolutional neural network model is integrated into the local processor of the AR binocular glasses. The recognition process includes insect species, wing type, quantity, and insect age identification, specifically: Determine whether the target is a rice planthopper, and then identify the rice planthoppers step by step. To determine the species of the target insect, the insect species is determined based on body color, markings and wing vein patterns, while the wing type is determined based on the ratio of wing length to body length. Among them, a wing length / body length ratio ≥ 0.8 is judged as a long-winged adult, and a wing length / body length ratio < 0.5 is judged as a short-winged adult; Determine the age of the target insect by identifying whether it is an adult or a nymph based on the relative size of its body outline and the completeness of its wing development; If the target is an adult insect, label it with the insect species and wing type, which are six types: M-WBPH, B-WBPH, M-BPH, B-BPH, M-SBPH, and B-SBPH. If the target is a nymph, its age can be initially determined based on its body shape and absolute size. Among them, a body length of 2-3mm is considered advanced age, and a body length of 0.1-2mm is considered young age; If it is an older nymph, it is labeled with a category based on the insect species and wing type, namely S-WBPH, S-BPH and S-SBPH. If it is a young nymph, a comprehensive judgment is made based on the precise size and subtle body color differences provided by the white disc image, and after eliminating the interference of impurities, the category label J-RPH is marked. After recognition is completed, record the bounding box coordinates and category label of each target in its respective sub-image; The recognition results of the sub-images are mapped back to the global coordinate system of the original image by an image coordinate mapping algorithm. The deduplication rule with an intersection-union ratio threshold of ≥0.3 is used to merge targets with overlapping positions and the same analog label in the global coordinate system, thereby obtaining the recognition results of each rice planthopper individual in the whole image and completing the automatic counting of each category and the calculation of the proportion of each insect age.

[0021] S4. Calculation of insect count per 100 clumps: The insect count per 100 clumps calculation module automatically calculates the insect count per 100 clumps according to the formula: insect count per 100 clumps = (total number of insects identified / number of clumps investigated) × 100, based on the total number of insects identified in step S3 and the number of clumps investigated. The number of clumps investigated is estimated through image analysis or input by the user. In this embodiment, the image analysis estimation method is adopted. S5. Data Storage and Real-time Display: The data storage and management module writes all data from this survey into the local non-volatile memory integrated in the AR binocular glasses for storage and automatically generates a survey report. The insect species and quantity, as well as the number of insects per hundred clusters, in the identification results are superimposed in real time on the AR binocular glasses' display screen through the binocular near-eye display module. This is done in a semi-transparent floating window overlaid on the real white disk image seen by the investigators, with a display effect as follows: Figure 4 ; The coordinate deviation between the superimposed position and the target coordinates of the rice planthopper in the real white disk image is 4 pixels; S6. Uploading and Visualizing Survey Results: After the survey concludes or when network access is available, the communication module will batch upload the survey report and all data to the cloud platform. The survey report can be viewed via a mini-program or a web-based computer. The mini-program display is shown below. Figure 5 The display effect on the computer web version is as follows Figure 6 .

[0022] Furthermore, the training method for a convolutional neural network model includes the following steps: The first step was to construct and label the dataset. Images of the white discs of the left and right eyes of multiple species of rice planthoppers were collected in multiple scenes using AR binocular glasses. The scenes covered the young nymphs (1st-2nd instar), old nymphs (3rd-5th instar), long-winged adults, and short-winged adults of white-backed planthopper, brown planthopper, and gray planthopper, covering different insect population densities, field lighting (sunny day, cloudy day, evening), and background interference (rice leaf fragments, soil particles). The LabelImg annotation software was used to draw bounding boxes based on the morphological features of each rice planthopper in the image, and a corresponding category label was set for each bounding box; the annotation accuracy was 2 pixels for the positional error of the bounding box. The second step is dataset preprocessing, which involves cutting the labeled white disk image into small sub-images of 512×512 pixels with an overlap rate of 35%, in order to increase the number of training samples and force the model to focus on learning local subtle features. The original white disk image resolution is 8000×6000, and the size of the small region sub-image is 512×512 pixels. Cutting it into sub-images can increase the proportion of young nymph targets in the sub-images by 3 times and effectively increase the number of training samples to 6 times that of the original samples. The third step is model training and validation. The YOLOv11 model is trained using the dataset from the second step, enabling the model to learn end-to-end. It jointly learns the color, texture, shape contour of a single frame image and the depth information and relative size information of the target obtained from binocular vision calculation. The texture structure includes the density of punctures on the adult wing surface and the distribution of setae on the nymph body surface. The model was optimized using stochastic gradient descent with 300 iterations and an initial learning rate of 0.01, which decayed by a factor of 10 every 100 iterations. Training was stopped when the loss function value on the validation set did not decrease for 10 consecutive iterations. Among them, the loss function uses CIoU loss superimposed with Focal loss to solve the problem of imbalance between bounding box regression and young nymph samples; The fourth step is model evaluation and optimization: the trained model is evaluated using a test set, with segmentation accuracy, recall and mAP50 as the core evaluation metrics; for young nymphs, a hard sample mining strategy is used for secondary fine-tuning until the recognition accuracy of each nymph age meets the preset requirements. Among them, the young nymphs are small in size and have indistinct morphological features, which can easily lead to low recognition accuracy. Therefore, a hard sample mining strategy is adopted during training to increase the training weight of young nymph samples. The final model achieved an accuracy of ≥95% for young nymphs, ≥97% for older nymphs, and ≥98% for adults on the test set.

[0023] like Figure 2 A smart identification system for rice planthoppers in the field based on AR binocular glasses, used to implement the aforementioned method, the system includes: The image acquisition module is used to acquire white disk images based on AR binocular glasses. The white disk images are RGB images. The binocular vision acquisition module includes a camera, a binocular near-eye display module, a positioning module, and a control module. The camera is mounted on the upper front of the AR binocular glasses and above the binocular near-eye display module, and is used to acquire RGB images of the white disc of rice planthoppers. The binocular near-eye display module is used to overlay and display the recognition results of the convolutional neural network model. The positioning module is used to obtain the geographical location information of the survey site, and the control module is used to realize the interaction between personnel and the system. The control module includes a voice recognition module and a touch module.

[0024] The binocular camera has a resolution of 4800w, a focal length of 5mm, and supports autofocus; the binocular near-eye display module has a resolution of 2560×1440 and a field of view of 80°; the positioning module includes a GPS positioning module with a positioning accuracy of ≤10m.

[0025] The local identification and calculation module includes a trained convolutional neural network model and a module for calculating the number of insects per 100 clusters. The convolutional neural network model and model file are stored in local non-volatile memory. The convolutional neural network model is trained to identify insect species, quantity, wing type and insect age. The module for calculating the number of insects per 100 clusters is used to automatically calculate the number of insects per 100 clusters and the proportion of each insect age. The module for calculating the number of insects per 100 clumps supports manual input of the number of clumps to be surveyed or automatic retrieval of the number of rice clumps in the identification results to determine the number of clumps to be surveyed. The local recognition and calculation module also includes an image segmentation processing module and a coordinate mapping and deduplication algorithm module. The image segmentation processing module is used to adjust the white disk image and cut it into multiple overlapping small sub-images of the same size as the model training stage, with an overlap rate of 30% to 50%. The coordinate mapping and deduplication algorithm module is used to map the recognition results back to the global coordinate system of the original image and delete duplicate recognized targets.

[0026] The data storage and management module automatically categorizes the survey data and securely writes it to local non-volatile memory in a structured format. Data is written according to a "date-plot-survey point" directory structure, and a log-structured file system ensures atomicity of data writing. Data is retained even in the event of an unexpected power outage. The data storage flowchart is shown below. Figure 3 The local non-volatile storage is 1TB of UFS 3.1; and the survey report is automatically generated based on the data; the data includes the original image, processed image, identification results, number of insects per hundred clusters, geographical location, survey timestamp, and surveyor ID; The data synchronization and display module includes a communication module, a cloud platform, and a display terminal. The communication module is used to upload locally stored data to the cloud platform. The communication module supports 4G, 5G, Wi-Fi, and Bluetooth and uses a breakpoint resume mechanism to upload data to the cloud platform. The cloud platform is used to receive, store, and manage data. The display terminal includes an APP, a mini-program, and a web terminal for visual display and sharing of survey reports.

[0027] In this embodiment, the various modules of the system are integrated and installed in the AR binocular glasses hardware body, which also includes a processor and a power module.

[0028] It should be noted that, in the description of this disclosure, unless otherwise expressly specified and limited, the terms "installation," "connection," and "joint" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this disclosure according to the specific circumstances.

[0029] The present invention has been described in detail above. However, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, any modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of rice planthoppers in the field based on AR binocular glasses, characterized in that, The method includes the following steps: S1. White Plate Image Acquisition: The investigators wear the AR binocular glasses and use a white plate to tap the rice clumps in the field. The acquisition is started by voice command or manual touch control module. The white plate image is acquired through camera to ensure that the image contains clear rice planthopper targets. S2. Image preprocessing: The image block processing module performs noise reduction and grayscale normalization on the white disk image, and cuts it into multiple overlapping small sub-images with the same size as the model training stage, with an overlap rate of 30%~50%. S3. Target Recognition and Counting: The trained convolutional neural network model is used to recognize the sub-images. This model is integrated into the local processor of the AR binocular glasses. The recognition process includes insect species, wing type, quantity, and age identification, specifically: Determine whether the target is a rice planthopper, and then identify the rice planthoppers step by step. To determine the species of the target insect, the insect species is determined based on body color, markings and wing vein patterns, while the wing type is determined based on the ratio of wing length to body length. Determine the age of the target insect by identifying whether it is an adult or a nymph based on the relative size of its body outline and the completeness of its wing development; If the target is an adult insect, label it with a category tag that combines the insect species and wing type; If the target is a nymph, its age can be initially determined based on its body shape and absolute size. If the target is an older nymph, label it with a category tag that combines the insect species and wing type; If the target is a young nymph, a comprehensive judgment is made based on the precise size and subtle body color differences provided by the white disc image, and the category label is marked after eliminating the interference of impurities. After recognition is completed, record the bounding box coordinates and category label of each target in its respective sub-image; The recognition results of the sub-images are mapped back to the global coordinate system of the original image through the image coordinate mapping algorithm. The deduplication rule with an intersection-union ratio threshold of ≥0.3 is used to merge targets with overlapping positions and the same analog label in the global coordinate system, so as to obtain the recognition results of each rice planthopper individual in the whole image and complete the automatic counting of each category and the calculation of the proportion of each insect age. S4. Calculation of insect count per 100 clumps: The insect count per 100 clumps calculation module automatically calculates the insect count per 100 clumps according to the formula: insect count per 100 clumps = (total number of insects identified / number of clumps investigated) × 100, based on the total number of insects identified in step S3 and the number of clumps investigated. The number of clumps investigated is estimated through image analysis or input by the user. S5. Data storage and real-time display: The data storage and management module writes all the data from this survey into the local non-volatile memory integrated in the AR binocular glasses for storage and automatically generates a survey report; The insect species and quantity in the identification results, as well as the number of insects per hundred clusters, are superimposed in real time on the display screen of the AR binocular glasses through the binocular near-eye display module, and superimposed on the real white disk image seen by the investigator in the form of a semi-transparent floating window; S6. Uploading and Visualizing Survey Results: After the survey is completed or when network access is available, the communication module will upload the survey report and all data in batches to the cloud platform; the survey report can be viewed via a mobile app or a web browser on a computer.

2. The intelligent identification method for rice planthoppers in the field based on AR binocular glasses according to claim 1, characterized in that, The category labels include long-winged adult white-backed planthopper, short-winged adult white-backed planthopper, older nymph white-backed planthopper, long-winged adult brown planthopper, short-winged adult brown planthopper, older nymph brown planthopper, long-winged adult gray planthopper, short-winged adult gray planthopper, older nymph gray planthopper, and young nymph gray planthopper.

3. The intelligent identification method for rice planthoppers in the field based on AR binocular glasses according to claim 1, characterized in that, The training method for the convolutional neural network model includes the following steps: The first step is dataset construction and annotation. White disc images of multiple species of rice planthoppers in multiple scenes are collected using the AR binocular glasses. Annotation software is used to draw bounding boxes based on the morphological features of each rice planthopper in the images, and a corresponding category label is set for each bounding box. The second step is dataset preprocessing, which cuts the labeled white disk image into multiple overlapping small region sub-images with an overlap rate of 30% to 50%, in order to increase the number of training samples and force the model to focus on learning local subtle features. The third step is model training and validation. The YOLOv11 model is trained using the dataset from the second step, enabling end-to-end learning. The model jointly learns the color, texture, shape contour of a single frame image and the hidden depth information and relative size information of the target in the white disk image. The model is optimized using stochastic gradient descent with 300 iterations and an initial learning rate of 0.01, which decays by a factor of 10 every 100 iterations. Training stops when the loss function value on the validation set does not decrease for 10 consecutive iterations. The fourth step is model evaluation and optimization: the trained model is evaluated using a test set, with segmentation accuracy, recall, and mAP as the core evaluation metrics; for young nymphs, a hard sample mining strategy is used for secondary fine-tuning until the recognition accuracy of each nymph age meets the preset requirements.

4. A smart identification system for rice planthoppers in the field based on AR binocular glasses, used to implement the method according to any one of claims 1 to 3, characterized in that, The system includes: The image acquisition module is used to acquire a white disk image based on the AR binocular glasses, wherein the white disk image is an RGB image; The local identification and calculation module includes a trained convolutional neural network model and a module for calculating the number of insects per hundred clusters. The convolutional neural network model is trained to identify insect species, quantity, wing type, and age. The module for calculating the number of insects per hundred clusters is used to automatically calculate the number of insects per hundred clusters and the proportion of each age. The data storage and management module is used to automatically classify the survey data and securely write it into local non-volatile memory in a structured format for storage, and automatically generate a survey report based on the data; the data includes original images, processed images, recognition results, number of insects per hundred clusters, geographical location, survey timestamp, and surveyor ID; The data synchronization and display module includes a communication module, a cloud platform, and a display terminal. The communication module is used to upload locally stored data to the cloud platform. The cloud platform is used to receive, store, and manage the data. The display terminal includes a mini-program and a web terminal for visually displaying and sharing survey reports.

5. The intelligent identification system for rice planthoppers in the field based on AR binocular glasses according to claim 4, characterized in that, The image acquisition module includes a camera, a binocular near-eye display module, a positioning module, and a control module; The camera is mounted on the upper front of the AR binocular glasses and above the binocular near-eye display module, and is used to capture RGB images of the white disc of the rice planthopper; the binocular near-eye display module is used to overlay and display the recognition results of the convolutional neural network model; the positioning module is used to obtain the geographical location information of the survey location; the control module is used to realize the interactive operation between the personnel and the system, and the control module includes a voice recognition module and touch buttons.

6. The intelligent identification system for rice planthoppers in the field based on AR binocular glasses according to claim 4, characterized in that, The local recognition and calculation module also includes an image segmentation processing module and a coordinate mapping and deduplication algorithm module; the image segmentation processing module is used to adjust the white disk image and cut it into multiple overlapping small sub-images of the same size as the model training stage, with an overlap rate of 30%~50%; the coordinate mapping and deduplication algorithm module is used to map the recognition results back to the global coordinate system of the original image and delete duplicate recognized targets.

7. A field rice planthopper intelligent identification system based on AR binocular glasses according to claim 5, characterized in that, The camera has a resolution of 4800w, a focal length of 5mm, and supports autofocus; the binocular near-eye display module has a resolution of 2560×1440 and a field of view of 80°; the positioning module includes a GPS positioning module with a positioning accuracy of ≤10m.

8. A field rice planthopper intelligent identification system based on AR binocular glasses according to any one of claims 4-7, characterized in that, The various modules of the system are integrated and installed in the AR binocular glasses hardware body, which also contains a processor and a power module.

Images