A Novel Fuzzy-Based Modified GAN and Faster RCNN for Classification of Banana Leaf Disease

摘要

In the global marketplace, agriculture plays an important role. However, diseases produced in plants mostly affect the financial system. Pest occurrence and climatic changes are the leading trouble that the banana plant leaf faces, which influences the quality of fruit and crop yields. To predict the leaf disease symptoms in an early stage, the automatic banana leaf disease detection technique is more significant. The traditional deep neural network methods used in banana plant leaf detection need more parameters and utilize complex architectures. To address such problems, modified generative adversarial networks-modified faster region-based convolutional neural networks with fuzzy (MGAN-MFRCNN with Fuzzy) is proposed for identifying banana leaf disease. The most effective framework in deep learning is GAN. The autoencoder is incorporated within the fundamental GAN design for developing a modified GAN model; it is used for data augmentation. To perform feature extraction from high-resolution networks, an MFRCNN is used in our proposed method. The improved and more accurate classification scores are obtained by combining fuzzy with the ensemble model. The experimentation is carried out using a banana plant leaf dataset in which the healthy leaf, xanthomonas-affected leaf, and sigatoka-affected leaf diseases are detected and classified. The proposed model reached 98% accuracy, 97% precision, 96% F-score, 98% specificity, and 98% sensitivity, which is better than other existing methods.