Potato Plant Disease Classification using Deep Features

Abstract

The potato is considerably most-liked  food crop and extensively cultivated on a global scale. The production of the crop is drastically decreased in terms of both quantity and quality due to fungal diseases, which pose a serious threat to the objectives of world's food security. In actuality, farmers make assumptions about these conditions by concentrating primarily on the leaves changing colour, which is typically dangerous owing to significant time commitment and subjective in nature. In these circumstances, the creation of automated computational models that statistically detect these diseases, even in their early stages, is crucial. This study suggests using digital photographs of potato leaves to automatically identify late and early blight illnesses using a hybrid model based on artificial intelligence  techniques. In this paper, two datasets are taken from publicly available PlantVillage database containing three classes labeled as healthy leaves, early blight, and late blight leaves. Deep features of both datasets are extracted using pre-trained ResNet18 model. Then the extracted feature vector of 512 features is passed to Support Vector Machine for building a model which identifies and classifies the diseased and healthy leaves. In comparison to a healthy potato leaf, the suggested framework has a test accuracy of 98.8% for identifying late and early blight syndromes. Additionally, the proposed methodology and the practices now in use have been contrasted. In this way, our recommended approach achieves good results with less resources and can be used efficiently and economically to classify the potato plant leaf diseases.