ResNet18-Based Intelligent System for Automated Potato Leaf Disease Detection and Classification

Abstract

This study leverages the ResNet18 deep learning model to classify potato leaves into three categories: Healthy, Early Blight, and Late Blight. Image dataset of high resolution was used in the study and the images were pre-processed by resizing, normalizing and by applying data augmentation to improve the predictive power of the model. To overcome this, the given data was divided into training, validation, and test datasets to maintain equal class distributions and to check the efficiency of the proposed approach. The ResNet18 model yielded an overall accuracy of 88%, with the precision, recall, and F1-scores being relatively even across all classes. Evaluation of confusion matrix and classification report brought about appreciation of the models’ performance, with only minor misclassifications mostly between similar diseases. This study demonstrates that ResNet18 is suitable for automated disease identification to enhance the management of diseases in farming. The future work involves collecting a larger and more diverse dataset, investigating the possibilities of utilizing advanced ensemble learning approaches and implementing the applications in real-world scenarios.

Topics & Concepts