Dual Branch Network by Fusing Standard Convolution and Dilated Convolution for Plant Leaf Diseases Classification

Most existing convolutional neural network (CNN)-based models for plant leaf disease and pest classification enhance recognition accuracy by stacking convolutional and pooling layers (i.e., increasing depth), but these architectures often suffer from high computational costs and excessive parameters. Given limited training samples, such large parameter volumes increase overfitting risks. To mitigate these issues, we introduce a novel dual branch network (DBNet) that synergistically combines standard and dilated convolutions. DBNet comprises two parallel branches: branch 1 extracts leaf features via a standard convolution followed by a dilated convolution, while branch 2 learns these different features. In addition, Dropout is applied between the two convolutions in every branch to minimize feature redundancy, and a sum fusion (SF) operation integrates multi-branch features. Experimental results confirm that DBNet achieves competitive classification performance while maintaining superior efficiency in computational complexity and parameter count compared to baseline models.