Hybrid optimization based deep learning network for multiclass plant leaf disease detection

Abstract

India is a rapidly growing country, and agronomy is the major pillar for the development of the country. However, crop cultivation faces numerous difficulties, which generates a massive loss in crop manufacturing. Plants are the significant part that supplies food and energy to human beings. Still, the disease in plants affects production and it causes economic losses. Moreover, it is considered as the cause for diminishing the quantity of crop products. These disorders mainly impact the leaves and create the diseases like spots, blight, and so on. Therefore, the finding of plant leaf disease is major part of the agricultural field. Moreover, early-stage disease detection is essential to obtain the effectual crop yield. Therefore, this thesis describes multiclass plant disease detection by two contributions. In the first contribution, the Geese Jellyfish Search Optimization (GJSO) with Deep Q-Network (DQN) termed as GJSO-DQN is utilized for identifying the disease. Moreover, the leaf is segmented via the Link-Net, and the GJSO is used to train it. The first level detects the plant category, and the second level detects the plant disease. The accuracy, False Positive Rate (FPR), and True Positive Rate (TPR) are considered to estimate the efficacy of the model. In the second contribution, the Fractional GJSO-based deep convolutional Neural Network (DCNN) called (FGJSO_ DCNN) is employed for the detection of plant leaf disease. The FGJSO- based Link-Net is utilized for leaf image segmentation, and the accuracy, FPR, and TPR are used to compute the performance. Here, also the first and second-level classifications are performed. Moreover, the implementations of these models are performed in the Python tool. In addition, the FGJSO_ DCNN model attained better accuracy, FPR, and TPR of 96.4%, 3.5%, and 95.5% in first level, and 96.3%, 3.3%, and 95.6% in second-level classifications...