Prediction of Diseases in Potato Plant using Pre-trained and Traditional Machine Learning Models

Abstract

Potato, among the most vegetables is commercially significant and well-known vegetable which is known for its high nutritional content and delicious flavor. India is one of the world’s leading producers of potato. Unfortunately, plant diseases in potato have been one of the causes of decreased production. So, it is necessary to detect them. Collecting images of plants diseases is a big challenge as it is a very time-consuming process. Often, we do not have sufficient data to train our deep learning models, so data augmentation techniques are used for increasing the dataset which lead to poor generalization. This study focuses on detecting whether the plant is healthy or diseased. In this proposed method, limited dataset is used for potato plant disease classification without using any data augmentation techniques. Popular pre-trained models — VGG16, InceptionResNetV2, ResNet50V2 are used for feature extraction and traditional machine learning algorithms — XGBoost, Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest are used as classifiers. From the study, it is observed that the combination of VGG16 model as a feature extractor and SVM as a classifier achieved the highest accuracy of 93% compared to rest of the combination of models and algorithms. The method proposed in this study can be used for potato plant disease detection with limited dataset.