SEAFEC: a spatial-edge adaptive convolution for multi-scale and boundary-aware plant disease and weed imagery

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Plant diseases and weeds pose significant threats to global crop production.
• Current deep learning methods struggle with multi-scale variations and blurred boundaries in agricultural imagery.

Purpose of the Study:

• To introduce SEAFEC (Spatial-Edge Adaptive Feature Enhancement Convolution), a novel convolutional module designed to enhance scale adaptivity and boundary precision.
• To address limitations in automated analysis of agricultural images, particularly for disease and weed identification.

Main Methods:

• SEAFEC utilizes a dual-branch design: the SCARF branch for dynamic receptive field adjustment and the MEFE branch for strengthening edge features.
• The module was evaluated on plant disease classification, corn leaf disease detection, and sugarcane-weed segmentation tasks.

Main Results:

• SEAFEC demonstrated consistent improvements across tasks: +1.8% accuracy in classification, +2.5% mAP in detection, and +3.4% mIoU in segmentation.
• Significant gains were observed in boundary-sensitive scenarios, indicating enhanced precision.

Conclusions:

• SEAFEC serves as a versatile enhancement module for agricultural image analysis.
• It offers a unified solution for scale-boundary challenges, supporting reliable disease diagnosis and precision weed management.