Deep feature extraction based cascading model for the classification of Fusarium stalk rot and charcoal rot disease in maize plant

Abstract

Maize stalk rot infection is one of the most important causes of production loss in some regions of India. However, it is difficult to diagnose maize disease on a wide scale due to a lack of human competency and limitations in standard image-recognition technologies. Fortunately, improvements in deep learning and machine learning technologies enable automatic disease identification. In this study, we proposed a 4-stage cascaded approach for the classification of Fusarium stalk rot and Charcoal rot in maize. In the proposed model, deep features are extracted by using three different convolutional neural network (CNN) models namely: GoogLeNet, ResNet18 and MobileNetV2 from the maize images which are preprocessed in the previous stage. The features extracted from these three CNN models are concatenated to benefit each CNN structure, and then the hybrid feature selection approach is utilized to choose the most relevant features with the lowest dimension to train the support vector machine (SVM) classifier. The findings of the experiment reveal that the proposed cascade model can classify the diseased as well as healthy images with an accuracy of 97.95%. Furthermore, the suggested model performed better in terms of specificity, accuracy, and F1-Score, with values of 0.98, 0.95, and 0.96, respectively. This study also presents a comparison of the performance of the proposed cascade model to that of several other relevant classification models.

Keywords

• Stalk rot
• Data augmentation
• CNN
• Transfer learning
• Hybrid feature selection
• Bidirectional search