Jute Leaf Disease Detection using ResNet50: A Deep Learning Approach for Precision Agriculture
Pratham Kaushik, Saniya Khurana
Abstract
Jute is one of the most important cash crops in many countries, and the health of this crop has a direct bearing on the agricultural and overall economy of those countries. This research applies a deep learning approach through ResNet50 to identify diseases of jute leaves. A dataset containing 1,820 jute leaf images was sourced from Kaggle, with labels divided into two classes: "Healthy" and "Diseased." The data was also rescaled, normalized, and augmented to increase the size of the data and reduce overfitting. The used ResNet50 model, pre-trained on ImageNet, was fine-tuned with a new classification head and reached an overall accuracy of 94% during evaluation time. Statistics, including precision, recall, and F1-score, were measured to have good performance with equal classification for both classes. A confusion matrix analysis also reveals that the model has low misclassification rates, as illustrated in the table below. This work demonstrates that deep learning can automatically identify jute leaf diseases, which can greatly aid in agricultural industry analysis and improve farming practices. The method in this paper is computationally efficient and can be applied in real-time systems. Possible future work can include the expansion of the model to other crop diseases and the implementation of the model in mobile or edge devices for on-field use, thus equipping farmers with fast and accurate diagnoses of diseases. This work is relevant to the field of smart agriculture and precision farming, as it is an emerging scientific area.