Plant Disease detection of Bell Pepper Plant Using Transfer Learning over different Models

Abstract

Bell Pepper, also called Capsicum, is a widely grown crop that suffers primarily from bacterial spot disease causing severe damage to the plant by dropping its leaves and fruits. The current methods to detect the disease are either through lab diagnosis, which takes quite an amount of time and may lead to further spreading of the disease throughout the crop, or an experienced eye. Focusing on this issue, this paper implements 3 Convolution Neural Network (CNN) architectures, namely, InceptionV3, ResNet50 and VGG16 via transfer learning technique to detect bacterial leaf spot disease in Bell Pepper Plant with the help of images from Plant Village dataset. These models were trained and tested over 2442 images of diseased and healthy leaves. The dataset images have also been augmented in various degrees to get better results. Further, fine tuning of the models by varying learning rates, batch size, epochs etc. has resulted in very high accuracies. VGG16 gives the best results with 99.72% accuracy and 0.998 AUC. ResNet50 showed 99.31% accuracy and 0.994 AUC whereas InceptionV3 resulted in the least of three with 95.77% accuracy and 0.953 AUC. These models are chosen due to their variable architectural differences and their performances are later compared using Accuracy, Loss, F1-score and AUC-ROC curve as performance metrics. The results in this paper show considerable improvement as compared to other research papers on detection of diseases in Bell Pepper plant. They can be helpful in the agriculture sector in timely detection of bacterial spot disease.

Topics & Concepts