High-accuracy crop disease and insect pest classification system

By designing a crop pest classification system including image preprocessing, feature extraction, object detection and output modules, the problem of insufficient robustness and generalization ability of the models in the prior art when processing unknown categories and sparse samples is solved, and high accuracy and generalization of crop pest identification is achieved.

CN119992197APending Publication Date: 2025-05-13南通西科瑞智能科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510083772.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-01-20
Publication Date
2025-05-13

AI Technical Summary

Technical Problem

Existing deep learning-based crop pest detection algorithms have poor model robustness and generalization capabilities when processing data with sparse numbers of unknown categories and samples, making it difficult to identify and process these data.

Method used

A high-accuracy crop pest classification system is designed, including image preprocessing module, feature extraction module, object detection module and output module. The image preprocessing module is trained through self-supervised learning, fine-grained features are extracted using modulation attention mechanisms, and an adaptive incremental classifier is designed through a hybrid relational mapping network and meta-learning method.

Benefits of technology

The system is able to extract more general and deeper feature representations, improving the generalization and recognition accuracy of the model, especially in the case of small sample data, and maintaining high accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119992197A_ABST
    Figure CN119992197A_ABST
Patent Text Reader

Abstract

The invention provides a high-accuracy crop disease and insect pest classification system. The system comprises an image preprocessing module, a feature extraction module, a target detection module and an output module which are connected in sequence, a modulation attention mechanism is introduced into the feature extraction module to calculate different weight scores for different positions of the disease and insect pest image so as to obtain more representative feature representation. The target detection module carries out training based on the feature data extracted by the feature extraction module, designs an adaptive increment classifier based on meta-training, comprises a hybrid relation mapping network, and carries out multi-stage training on the hybrid relation mapping network by using a pseudo-increment plot training method based on meta-learning to extract inductive deviation during meta-training, so as to obtain a target detection result. The task is popularized to an increment task; and designing a hybrid relation mapping network based on Transform, wherein the hybrid relation mapping network comprises a prototype self-mapping network, a feature representation cross-mapping network and a prototype cross-mapping network. According to the method, high model accuracy can be kept in a small amount of sample data.
Need to check novelty before this filing date? Find Prior Art