All-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT

Abstract

The invention discloses a multi-task crop health monitoring and early warning platform based on an EffiDIO-ViT model, and the platform integrates an EffiDIO-ViT multi-task learning model, a geographic information system (GIS) module, a disease trend prediction module based on XGBoost, an interaction module and a data storage module, and achieves the precise recognition, feature extraction and trend prediction of diseases and pests in a cooperative manner. According to the platform, a multi-illumination all-weather data set generated based on Blender simulation is adopted, the detection capability of an EffiDIO-ViT model under a complex illumination condition is effectively improved, and a powerful technical support is provided for all-weather crop health monitoring. The XGBoost-based disease trend prediction module is composed of a disease analysis unit and an XGBoost prediction unit, the disease analysis unit takes visual information output by an EffiDINO-ViT model as a core, extracts classification results, segmentation regions and related features of diseases and pests, optimizes feature expression through conditional position encoding, and performs feature integration in combination with a multi-layer sensor (MLP).

Description

Technical Field

[0001] The present invention relates to the technical field of crop monitoring, and in particular to an all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT. Background Art

[0002] As a major agricultural country, China's agricultural development is closely related to people's livelihood and well-being. It emphasizes accelerating the construction of intelligent industrial chains in rural areas, implementing efficient and intelligent management, and speeding up the process of rural modernization to promote the rapid development of agriculture in the direction of modernization and intelligence.

[0003] Currently, most smart agriculture solutions on the market focus on the identification and follow-up education of pests and diseases. Only a few products provide value-added remote monitoring, disease prediction, and intelligent care services. The existing classification mechanism has problems with low credibility and incomplete category coverage in pest identification. The classification process relies on additional information such as temperature, humidity, and geographic location. It is difficult to effectively classify leaf diseases and pests at night. A few products support leaf disease detection, but are limited to providing bounding box identification of diseased areas and lack the function of detailed segmentation of lesions. Most current products do not support pest detection and do not provide the function of segmenting pest areas. The disease prediction function usually uses a linear prediction model, resulting in limited prediction accuracy. The linear prediction model often faces the challenge of limited prediction accuracy because it assumes that the disease development follows a linear trend. This model structure is difficult to capture the dynamic changes and variability of disease development when dealing with complex, nonlinear disease development processes. Summary of the invention

[0004] The main purpose of the present invention is to provide an all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT, which can effectively solve the technical problems in the background technology.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT includes an EffiDINO-ViT model, an all-weather data set, a geographic information system integration module, a disease trend prediction module, an interaction module and a storage module. The disease trend prediction module consists of a disease analysis unit and an XGBoost prediction unit. The disease analysis unit takes the visual information output by the EffiDINO-ViT model as the core to extract the classification results, segmentation areas and related features of diseases and pests; the XGBoost prediction unit uses the extraction results of the analysis unit combined with other relevant data to predict the development trend of the disease and the possible diseased ear rate.

[0007] Preferably, the disease analysis unit captures crop image data through a fixed-point camera, classifies and segments pest and disease targets based on the EffiDINO-ViT model, accurately identifies diseased areas, and extracts relevant features for quantifying the severity of the disease.

[0008] Preferably, the all-weather dataset uses synthetic data or semi-supervised learning to simulate low-light scenes, so that the model can maintain high detection and segmentation accuracy under never-seen low-light conditions. On the one hand, the dataset uses a synthetic dataset and uses Blender software to simulate and generate a crop segmentation dataset under multiple lighting conditions. On the other hand, data enhancement is used to simulate multiple lighting for real images, ensuring that the model can maintain high detection and segmentation accuracy under various environmental conditions, providing a new perspective for the application of deep learning models under complex lighting conditions.

[0009] Preferably, the disease trend prediction model based on XGBoost is based on the Boost regression method, especially using the XG-Boost algorithm, to construct a learnable disease prediction model. This model will conduct in-depth analysis based on the monitored visual information, such as the ratio of disease / pest area to leaf area, the type and quantity of pests and diseases, and other data, and predict the development trend of the disease to help farmers respond in time; the Boost regression method deeply analyzes the disease data and develops an accurate mathematical model to predict the development trend of the disease and the diseased ear rate. Through this method, it is possible to introduce human intervention factors based on the comprehensive existing data to predict the development trend of crops in the future.

[0010] Preferably, the interactive module enables farmers to access the health status and warning information of crops anytime and anywhere through the APP and web interface. Through real-time data display, historical trend analysis and customized warning settings, farmers can quickly obtain key information and make reasonable decisions, helping farmers understand the best time for pest control and disease prevention, and further enhancing the comprehensive value of this smart agriculture solution.

[0011] Preferably, the synthetic data is divided into two parts: a validation set rendered using the Eevee engine, including 1,500 images, and a training set rendered using the Cycles engine, including 500 images. This division of labor ensures the diversity, efficiency and quality of the training and validation data, and provides a comprehensive learning and validation environment for the model. Through this method, the lighting conditions in the data set can be precisely controlled, thereby better evaluating and optimizing the performance of the segmentation algorithm under different lighting conditions.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] In the present invention, by setting up an all-weather data set, using synthetic data or semi-supervised learning to simulate low-light scenes, the model can still maintain high detection and segmentation accuracy under unseen low-light conditions. On the one hand, the data set uses a synthetic data set, and the project team uses Blender software to simulate and generate a crop segmentation data set under multiple lighting conditions; on the other hand, data enhancement is used to simulate multiple lighting conditions for real images. This strategy ensures that the model can maintain high detection and segmentation accuracy under various environmental conditions, providing a new perspective for the application of deep learning models under complex lighting conditions.

[0014] By setting up a geographic information system integration module, conditional position coding is used in the key link of feature extraction, and unified integration and optimization are carried out through the MLP head to adapt to the structure of the encoder. This core technical route integrates the unified processing mechanism of conditional position coding, which not only works in coordination with the structure of the encoder, but also greatly improves the efficiency of information processing;

[0015] By setting up a disease trend prediction model based on XGBoost, based on the Boost regression method, especially using the XG-Boost algorithm, a learnable disease prediction model was constructed. This model will conduct in-depth analysis based on the monitored visual information, such as the ratio of disease / pest area to leaf area, the type and quantity of pests and diseases, and other data, and predict the development trend of the disease to help farmers respond in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a structural schematic diagram of the all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT of the present invention;

[0017] Figure 2 This is a feature extraction backbone network diagram of the all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT of the present invention;

[0018] Figure 3 This is a network architecture diagram for disease detection and segmentation of the all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT of the present invention. DETAILED DESCRIPTION

[0019] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0020] like Figure 1-3The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT is shown, and the all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT includes an EffiDINO-ViT model, an all-weather data set, a geographic information system integration module, a disease trend prediction module, an interaction module and a storage module. The disease trend prediction module consists of a disease analysis unit and an XGBoost prediction unit, wherein the disease analysis unit takes the visual information output by the EffiDINO-ViT model as the core to extract the classification results, segmentation areas and related features of pests and diseases; the XGBoost prediction unit uses the above visual information in combination with external data (such as climate and geographic information) to predict disease trends, effectively improving the prediction accuracy and reliability. In the key link of feature extraction of the disease analysis and prediction unit, conditional position coding is adopted, and unified integration and optimization are performed through the MLP head to adapt to the structure of the encoder. This core technical route integrates the unified processing mechanism of conditional position coding, which not only works in coordination with the structure of the encoder, but also greatly improves the efficiency of information processing.

[0021] The disease analysis and prediction unit uses fixed-point cameras to capture crop image data and accurately identify pests and diseases. The image segmentation unit segments the diseased area and then conducts in-depth quantitative analysis on these segmented areas to estimate the severity of the disease. The all-weather data set uses synthetic data or semi-supervised learning to simulate low-light scenes, so that the model can still maintain high detection and segmentation accuracy under unprecedented low-light conditions. On the one hand, the data set uses a synthetic data set and uses Blender software to simulate and generate a crop segmentation data set under multiple lighting conditions. On the other hand, data enhancement is used to simulate multiple lighting for real images, ensuring that the model can maintain high detection and segmentation accuracy under various environmental conditions, providing a new perspective for the application of deep learning models under complex lighting conditions. The disease trend prediction model based on XGBoost is based on Boost regression. The Boost regression method, especially the use of the XG-Boost algorithm, has constructed a learnable disease prediction model. This model will conduct in-depth analysis based on the monitored visual information, such as the ratio of disease / pest area to leaf area, the types and quantity of pests and diseases, and other data, and predict the development trend of the disease to help farmers respond in time; the Boost regression method deeply analyzes the disease data and develops an accurate mathematical model to predict the development trend of the disease and the diseased ear rate. Through this method, on the basis of comprehensive existing data, human intervention factors can be introduced to predict the development trend of crop diseases and pests in the future. The prediction method takes into account the direct data provided by the model, and also integrates the influence of external variables, such as climate change and adjustment of agricultural time, so that the prediction results are more in line with the actual situation, which not only improves the accuracy of the prediction, but also enhances the applicability of the model. The goal of this study is to develop an early warning system that can provide agricultural practitioners and researchers with key information on the potential spread of diseases in a timely manner, so that they can develop effective prevention and control strategies in advance; the interactive module, through the APP and web interface, enables farmers to access the health status and warning information of crops anytime and anywhere. Through real-time data display, historical trend analysis and customized warning settings, farmers can quickly obtain key information and make reasonable decisions, helping farmers understand the best time for disease and pest control, and further enhance the comprehensive value of this smart agriculture solution; the synthetic data is divided into two parts: a validation set rendered using the Eevee engine, including 1,500 images, and a training set rendered using the Cycles engine, including 500 images. This division of labor ensures the diversity, efficiency and quality of training and verification data, and provides a comprehensive learning and verification environment for the model. In this way, the lighting conditions in the data set can be precisely controlled, so as to better evaluate and optimize the performance of the segmentation algorithm under different lighting conditions.

[0022] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. All-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT, characterized by: It includes an EffiDINO-ViT model, an all-weather data set, a geographic information system integration module, a disease trend prediction module, an interaction module and a storage module. The disease trend prediction module is composed of a disease analysis unit and an XGBoost prediction unit. The disease analysis unit takes the visual information output by the EffiDINO-ViT model as the core to extract the classification results, segmented areas and related features of pests and diseases; the XGBoost prediction unit uses the extraction results of the analysis unit in combination with other relevant data to predict the disease development trend and possible diseased ear rate.

2. The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT according to claim 1, characterized in that: The disease analysis unit captures crop image data through a fixed-point camera, classifies and segments pest and disease targets based on the EffiDINO-ViT model, accurately identifies diseased areas, and extracts relevant features for quantifying the severity of the disease.

3. The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT according to claim 2 is characterized by: The all-weather dataset uses synthetic data or semi-supervised learning to simulate low-light scenes, so that the model can still maintain high detection and segmentation accuracy under never-seen low-light conditions. On the one hand, the dataset uses a synthetic dataset and uses Blender software to simulate and generate a crop segmentation dataset under multiple lighting conditions. On the other hand, data enhancement is used to simulate multiple lighting for real images, ensuring that the model can maintain high detection and segmentation accuracy under various environmental conditions, providing a new perspective for the application of deep learning models under complex lighting conditions.

4. The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT according to claim 3 is characterized by: The disease trend prediction model based on XGBoost is based on the Boost regression method, especially the XG-Boost algorithm, to construct a learnable disease prediction model. This model will conduct in-depth analysis based on the monitored visual information, such as the ratio of disease / pest area to leaf area, the type and quantity of pests and diseases, and other data, and predict the development trend of the disease to help farmers respond in time; the Boost regression method deeply analyzes the disease data and develops an accurate mathematical model to predict the development trend of the disease and the diseased ear rate. Through this method, it is possible to introduce human intervention factors on the basis of comprehensive existing data to predict the development trend of crop diseases and pests in the future.

5. The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT according to claim 4 is characterized by: The interactive module enables farmers to access the health status and warning information of crops anytime and anywhere through the APP and web interface. Through real-time data display, historical trend analysis and customized warning settings, farmers can quickly obtain key information and make reasonable decisions, helping farmers understand the best time for pest control and further enhance the comprehensive value of this smart agriculture solution.

6. The all-weather multi-task crop health monitoring and early warning platform based on EffiDINO-ViT according to claim 5, characterized in that: The synthetic data is divided into two parts: a validation set rendered using the Eevee engine, including 1,500 images, and a training set rendered using the Cycles engine, including 500 images. This division of labor ensures the diversity, efficiency and quality of the training and validation data, providing a comprehensive learning and validation environment for the model. Through this method, the lighting conditions in the dataset can be precisely controlled, thereby better evaluating and optimizing the performance of the segmentation algorithm under different lighting conditions.

Images