A Novel Feature Extraction and Siamese Zeiler and Fergus Forward Taylor Network-Based Rice Plant Leaf Disease Detection

Abstract

Rice leaf disease affects the leaves of the rice plant that are caused by fungi, bacteria or viruses. Leaf disease leads to yellowing, wilting or lesions on the leaves, which affects photosynthesis and minimises crop production. General rice leaf diseases include rice blast, bacterial blight, and leaf smut, which reduce food production and the economic stability of farmers. Hence, rice plant leaf disease detection is an important aspect, which ensures healthy crop yields. Many methods have been proposed for rice plant leaf disease detection, but they did not fully handle the variability in disease symptoms. Therefore, Siamese Zeiler and Fergus Forward Taylor Network (S-ZFFTNet) is developed for rice plant leaf disease detection. First, leaf disease images are collected from the rice leaf bacterial and fungal disease dataset and denoised by anisotropic diffusion. The plant leaf is segmented by conditional Generative Adversarial Network (cGAN). Then, the segmented image is augmented by rotation, colour change, and scaling factor. Then, Fuzzy Local Binary Patterns (FLBP) with wavelet transform features are excerpted from an augmented image. In the rice plant leaf disease detection phase, a new hybrid S-ZFFTNet is utilised, which is the unification of the Siamese Convolutional Neural Network (SCNN), Zeiler and Fergus Network (ZF-Net), and Taylor's series. The results acquired by S-ZFFT-Net are 92.654% of accuracy, 94.654% True Positive Rate (TPR), 91.757% True Negative Rate (TNR), 90.866% precision, and 92.721% F1-score for k fold value 8.