DCCAM-MRNet: Mixed Residual Connection Network with Dilated Convolution and Coordinate Attention Mechanism for Tomato...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Tomato crops are susceptible to bacterial and viral diseases, impacting yield and quality.
• Current machine vision methods struggle with detecting subtle disease features on tomato leaves due to complex environments and small lesion sizes.

Purpose of the Study:

• To develop a novel and accurate method for detecting tomato leaf diseases using advanced image recognition.
• To overcome the limitations of existing machine vision approaches in identifying inconspicuous disease symptoms.

Main Methods:

• Implemented an integration nonlocal means (INLM) filtering algorithm to reduce image noise.
• Developed a novel deep learning network, DCCAM-MRNet, using ResNeXt50 as the backbone.
• Incorporated Dilated Convolution (DC) for expanded perceptual fields and coordinate attention (CA) for precise localization of disease spots.
• Utilized a mixed residual connection (MRC) technique combining RS-Block and TRS-Block to enhance accuracy and reduce model size.

Main Results:

• The proposed DCCAM-MRNet achieved a classification accuracy of 94.3%.
• The network demonstrated superior performance compared to existing tomato leaf disease detection networks.
• The model size was reduced by 0.11M parameters compared to the ResNeXt50 backbone.

Conclusions:

• The combination of INLM filtering and the DCCAM-MRNet model presents a successful strategy for effective tomato disease identification.
• The developed method offers a significant advancement in precision agriculture for disease management in tomato cultivation.