Plant Disease Detection and Severity Assessment Using Image Processing and Deep Learning Techniques

Efficient plant disease detection and severity assessment are crucial not just for agricultural purposes but also for global health, economics, as well as ecological sustainability. With the help of innovative computational techniques, we need to build resilient agricultural systems for a sustainable future. In this paper, firstly, the authors implemented four distinct image enhancement techniques. Based on the results, the technique with the best accuracy measures was selected for further implementation. Next, six CNN architectures namely AlexNet, ResNet18, ResNet50, ResNet101, SqueezeNet, and Inception V3 were implemented on an original image dataset constituting tomato early blight leaf images. Thereafter, image processing was performed on the images in order to enhance their quality and size. For disease detection, AlexNet, SqueezeNet, ResNet18, ResNet50, ResNet101, and Inception V3 achieved an accuracy of 96.43%, 97.32%, 99.11%, 99.55%, 97.32%, and 98.66%, respectively. Next, the images were divided into classes of disease severity, namely healthy, early, middle, and late, for which the accuracies achieved by all CNNs ranged between 66.88% and 78.98%. Next, the six CNN models were used only for feature extraction and SVM was applied for classification. The best accuracy of 82.80% was achieved via ResNet101 architecture. A similar implementation was done after performing segmentation on the images in the dataset. © 2023, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.