Harnessing the Potato leaf disease detection process through proposed Conv2D and resnet50 deep learning models

Abstract

This paper endeavours to implement the image processing technique for finding out the potato leaf disease. Total two categories of leaf disease—early blight and late blight—have been taken into consideration for the data processing along with the healthy leaf. The total number of cases is 3251 for the transfer learning process in the leaf disease detection process. Before incorporating the data in the deep learning model, image size has been converted to (224, 224) along with the normalization of pixel values. Total two deep learning models- CNN and Resnet50—have been implemented to perform the potato leaf disease detection process with 20 epochs each. From the analysis of the results, it is found that reset50 has successfully tracked the validation accuracy with 97% overall, while the CNN has given 76% of validation accuracy. Finally, the classification report and the confusion matrix for each of the models have been produced to see the overall performance for the potato leaf disease detection process. And, the training loss and the validation loss have been documented in terms of graphical order of representation for deeper understanding of the model performance.