Maize Leaf Disease Recognition Based on Improved Convolutional Neural Network ShuffleNetV2.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Maize diseases pose significant management challenges.
• Traditional identification methods lack accuracy and are difficult for mobile deployment.
• Need for efficient, accurate, and mobile-compatible disease detection systems.

Purpose of the Study:

• To develop an accurate and compact corn leaf disease recognition model for mobile devices.
• To improve upon existing convolutional neural network models for maize disease identification.
• To facilitate precision agriculture through automated disease detection.

Main Methods:

• Proposed a novel model, SNMPF, based on ShuffleNetV2 convolutional neural network.
• Integrated a max pooling layer for down-sampling to enhance feature extraction and generalization.
• Incorporated the Sim AM attention mechanism to improve feature expression in complex backgrounds.

Main Results:

• The SNMPF model achieved a high recognition accuracy of 98.40%.
• The model size is compact at only 1.56 MB, suitable for mobile applications.
• Demonstrated superior performance compared to EfficientNet, MobileViT, EfficientNetV2, RegNet, and DenseNet.

Conclusions:

• The SNMPF model offers a highly accurate and efficient solution for maize leaf disease recognition.
• The model's compact size and high accuracy support automated detection in natural field conditions.
• Results provide scientific guidance for disease prevention and advance precision agriculture practices.