Optimized deep learning model for plant disease detection using leaf images

Abstract

Accurate and timely plant disease detection is very much essential in crop health and agricultural yield. The work below describes an improved deep learning model for the classification of plant diseases through images. In doing so, it considers a dataset of 124,636 raw images taken against 37 categories involving healthy and diseased plants. All of the models that were based on VGG16, VGG19, InceptionV3, MobileNetV2, ResNet50, and a hybrid model that integrated ResNet152 with DenseNet201 have been evaluated. Among the tested models, the proposed hybrid model attained the best result in terms of accuracy and robustness. Also utilising LRA (Learning Rate A) reduced the number of iterations without affecting the accuracy. This approach has significantly enhanced plant disease classification. Therefore, this approach is efficient and extendable for automatic disease detection. Our work depicts how DL can prominently support precision agriculture by enabling early detection and prevention of agricultural damage.