Maize Small Leaf Spot Classification Based on Improved Deep Convolutional Neural Networks with a Multi-Scale Attention Mechanism

Maize small leaf spot (Bipolaris maydis) is one of the most important diseases maize. The severity disease cannot be accurately identified, cost pesticide application increases every year, and agricultural ecological environment polluted. Therefore, in order to solve this problem, study proposes a novel deep learning network DISE-Net. We designed dilated-inception module instead traditional inception for strengthening performance multi-scale feature extraction, then embedded attention learn importance interchannel relationships input features. In addition, dense connection strategy used model building strengthen channel propagation. paper, we constructed data set maize spot, including 1268 images four grades healthy leaves. Comparative experiments show that DISE-Net with test accuracy 97.12% outperforms classical VGG16 (91.11%), ResNet50 (89.77%), InceptionV3 (90.97%), MobileNetv1 (92.51%), MobileNetv2 (92.17%) DenseNet121 (94.25%). Grad-Cam visualization also shows able pay more key areas making decision. results showed was suitable classification field.