Agro Vision: Utilizing Drones and Deep Learning for Precise Plant Disease Detection in Horticulture Via Image Analysis

Applications of advanced techniques of deep learning (DL) transfer learning for plant disease prediction have been gaining significant momentum during the last few years, especially within agriculture, where timely diagnosis may be all-important for crop management. This study proposes a hybrid transfer learning model combining VGG-19 and ResNet-50 architectures for predicting diseases in various leaves. The deep convolutional layers with the capability to extract highly detailed features are inherited from VGG-19. In contrast, ResNet-50 uses residual learning to overcome the vanishing gradient problem and enhances feature representation. Thus, the hybrid model proposed here takes the strengths of each network architecture to build better diagnostics accuracy and robustness. The model would classify different disease states of its kind with good accuracy by using transfer learning and fine-tuning techniques over a dataset of images of plant leaves, thus providing an application that could help in the early detection of such diseases. This combination-VGG-19 and ResNet-50-would help in achieving strong feature extraction and superior performance with respect to minute symptoms of diseases, which would lead towards better agricultural practices. The proposed transfer learning techniques have been tested on plant disease datasets. It is expected that the system will be able to reach an accuracy rate of above 97%.