Research On Maize Disease Identification Methods In Complex Environments Based On Cascade Networks And Two-Stage...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate maize disease identification is challenging due to complex environmental contexts in natural images.
• Existing methods struggle with image complexity, potentially misinterpreting non-disease elements as symptoms.

Purpose of the Study:

• To develop and evaluate a novel method for accurate maize leaf disease identification and classification in complex environments.
• To improve the efficiency and speed of training machine learning models for crop disease detection.

Main Methods:

• A cascade network combining a Faster R-CNN leaf detector (LS-RCNN) with an attention mechanism and a CNN disease classifier (CENet).
• Implementation of a two-stage transfer learning strategy for training the CENet model on complex-context disease images.
• Classification of four maize leaf types: Cercospora leaf spot, Common rust, Northern Leaf Blight, and Healthy.

Main Results:

• The proposed method achieved a high F1-score of 99.70% for identifying four types of maize leaves.
• Demonstrated superior performance compared to popular Convolutional Neural Network (CNN) models and other existing methods.
• Exhibited a significant advantage in terms of model training speed.

Conclusions:

• The developed cascade network and transfer learning approach effectively addresses the challenge of maize disease identification in complex environments.
• The method offers a robust and efficient solution for accurate crop disease classification, with potential applications for other crop varieties.
• The findings highlight the significance of advanced deep learning techniques for agricultural applications.