Comparative result analysis of cauliflower disease classification based on deep learning approach VGG16, inception v3, ResNet, and a custom CNN model

Abstract

Out of many threats, plant diseases are the major ones to agriculture globally. They can drastically reduce productivity and lead to substantial economic losses. Traditional disease detection methods around these areas are often time-consuming, costly, and less effective, leading to the exploration of advanced techniques such as deep learning. In this study, we compared the results of three different deep learning approaches, namely VGG16, Inception v3, ResNet, and a custom CNN model for the detection of plant diseases in the context of tropical regions. To evaluate the performance of each approach, we used a dataset consisting of images of cauliflower plant diseases commonly found in countries like Bangladesh, India, and others. We trained each model using a transfer learning approach, where we used pre-trained models initially trained on the VegNet dataset on various train-validation splits. Various evaluation metrics were used to conduct this study: accuracy, precision, loss, recall, and F1 score. The ResNet50 model performed the best with an accuracy of 90.85 %, followed by our proposed model with an accuracy of 89.04 %. The findings suggest that deep learning approaches, especially Resnet50, and the proposed model can effectively detect diseases in tropical regions. The study's results suggest that using advanced technologies, such as deep learning, can significantly enhance the effectiveness of disease detection and control, leading to improved agricultural productivity and food security.