Geese jellyfish search optimization trained deep learning for multiclass plant disease detection using leaf images

Abstract

Accurate and early detection of plant disease is significant for stable and proper agriculture and also for preventing the unwanted waste of financial and other possessions. Hence, a new technique is devised in this work, where geese jellyfish search optimization trained deep learning is used for multiclass detection of plant disease utilizing plant leaf images. At first, the input leaves of the plant image acquired from the database are pre-processed utilizing the Kalman filter. Then, the plant leaf segmentation is done by LinK-Net, where the training function of LinK-Net is processed by the proposed geese jellyfish search optimization, which is formed using wild geese migration optimization and jellyfish search optimizer. Then, image augmentation is carried out and then the feature extraction is done. Consequently, the classification of plant leaf type is processed, which is employed by Deep Q-Network (DQN), which is structurally adapted by the proposed geese jellyfish search optimization. At last, multi-label plant leaf disease is detected based on DQN. Moreover, the proposed geese jellyfish search optimization based DQN obtains an accuracy of 89.44%, true positive rate of 90.18%, and false positive rate of 10.56% respectively.