A Convolutional Neural Network-Based Plant Pesticide Recommendation System for Rural Areas

Problems with crop production are widespread in India and have a significant impact on the country's economy, particularly on rural farmers. Depending on the state of the leaf, farmers may predict the amount and quality of their harvest in advance, making the leaf an essential component of crop management. In this study, we provide a system that uses Tensor Flow technology for preprocessing and feature extraction of plant village dataset leaf pictures, then a convolution neural network for illness classification and pesticide recommendation. Our solution primarily makes use of two processes: deep learning and an Android app that makes use of Java Web Services. We trained our model using a Convolution Neural Network with five, four, and three layers, and we interfaced it with a user-friendly Android app and JWS. We found that a 5-layer model utilizing tensor flow attained the best accuracy at 95.05% after 15 epochs, and the maximum validation accuracy at 89.67% after 20 epochs.