Segmentation-based lightweight multi-class classification model for crop disease detection, classification, and severity assessment using DCNN

Leaf diseases in Zea mays crops have a significant impact on both the calibre and volume of maize yield, eventually impacting the market. Prior detection of the intensity of an infection would enable the efficient allocation of treatment resources and prevent the infection from spreading across the entire area. In this study, deep saliency map segmentation-based CNN is utilized for the detection, multi-class classification, and severity assessment of maize crop leaf diseases has been proposed. The proposed model involves seven different maize crop diseases such as Northern Leaf Blight Exserohilum turcicum, Eye Spot Oculimacula yallundae, Common Rust Puccinia sorghi, Goss's Bacterial Wilt Clavibacter michiganensis subsp. nebraskensis, Downy Mildew Pseudoperonospora, Phaeosphaeria leaf spot Phaeosphaeria maydis, Gray Leaf Spot Cercospora zeae-maydis, and Healthy are selected from publicly available datasets obtained from PlantVillage. After the disease-affected regions are identified, the features are extracted by using the EffiecientNet-B7. To classify the maize infection, a hybrid harris hawks' optimization (HHHO) is utilized for feature selection. Finally, from the optimized features obtained, classification and severity assessment are carried out with the help of Fuzzy SVM. Experimental analysis has been carried out to demonstrate the effectiveness of the proposed approach in detecting maize crop leaf diseases and assessing their severity. The proposed strategy was able to obtain an accuracy rate of around 99.47% on average. The work contributes to advancing automated disease diagnosis in agriculture, thereby supporting efforts for sustainable crop yield improvement and food security.