Classification of Tobacco Leaf Pests Using VGG16 Transfer Learning

Some of the attacking pest species of tobacco leaves were witnessed only after carrying out the initial fermentation. Infestation adversely affects the quality of tobacco leaves. Infested leaves must be separated from healthy ones to maintain quality. The sorting process, which is prone to human errors, tends to be manual. As part of a larger study, we aimed to automate the identification and classification of tobacco leaves afflicted by many pest attacks. Convolutional Neural Network (CNN) was one of the recent classification techniques proposed in this paper using the renowned VGG16 architecture. Training VGG16 from scratch, that is to say, random initialization of weights, could take an extremely long time. This is why we set initial weights via transfer learning, so as to increase accuracy and speed the training time. We have now achieved an accuracy previously unattained in other research using only VGG16 and transfer learning for single class of the disease. Some experiments were carried out in order to exhaustively combine number of learnable parameters and types of optimizers to reach the optimal results. The configuration was subsequently verified with success at a really impressive-high accuracy output.

Keywords: Tobacco leaf classification, pest detection, VGG16, transfer learning, convolutional neural networks, plant disease identification, image classification, deep learning, agricultural automation, quality control.