A Blockchain-Based Hybrid Hunger Game Search Archimedes Optimization Enabled Deep Learning for Multiclass Plant Disease Detection Using Leaf Images

Abstract

Plants are susceptible to a wide range of diseases when they are growing. One of the crucial difficulties in agriculture is the earlier finding of plant diseases. If the diseases are not detected at the beginning, it may have an undesirable effect on the entire production. To avoid these issues, a blockchain-based hybrid optimized deep learning (DL) approach is devised in this work. The plant leaf images are stored in the blockchain network and the noise level of the images is minimized by the Kalman filter. In image segmentation, the Deep Joint segmentation technique is employed to segment the disease-affected portion of the image. The position and color augmentation are carried out to enhance the size and clarity of the image. Moreover, the statistical and speeded-up robust features (SURF) are extracted in the feature extraction stage. In the first level classification process, the developed hunger game search Archimedes optimization (HGSAO) enabled SpinalNet is employed for classifying the plant type and the second level classification is carried out for multiclass disease identification using the proposed HGSAO optimized SpinalNet. Moreover, the proposed HGSAO with SpinalNet outperformed the accuracy of 0.972, True positive rate (TPR) of 0.963, true negative rate (TNR) of 0.951, false negative rate (FNR) of 0.936 and false positive rate (FPR) of 0. 942.