A Multi-stage Deep Model for Crop Variety and Disease Prediction

Abstract

One of the most important and crucial professions in the world is agriculture. It is estimated that various kinds of pests like insects, weeds, animals, and diseases cause crop yield losses of 20–40%. This raises the need for detection of the types and severity of diseases at early stages, thus minimizing the usage of fertilizers and the chances of the producing healthy and higher crop yields. Therefore, it is now vital to raise the quality of agricultural products. Artificial intelligence has a great deal of potential to be demonstrated as an effective instrument that can assist agriculture in managing the growing complexity of modern agriculture. Especially doing agriculture in large scale can benefit from the intelligent systems. Till now the systems for identifying and classifying crop varieties and crop diseases have been undertaken as two different problems and the deep learning solutions developed are isolated. We have initiated and directed our work towards developing a multi-stage deep learning model which will predict the crop varieties along with the crop diseases in a composite manner. The intention was to allow integration of learning models for achieving multiple tasks. We have used ResNet18 and VGG19 for learning and classifying the crop varieties and crop diseases. We have performed experimentation with varying hyperparameter setup and the results were analyzed. The suggested model provides effectiveness by correctly detecting and recognizing crop varieties and diseases with a top-1 training accuracy of 94 .91% and a top-1 validation accuracy of 90.9%.