Enhancing Plant Leaf Disease Prediction Through Advanced Deep Feature Representations: A Transfer Learning Approach

Plant diseases present a considerable threat to the farming industry, causing significant economic losses by reducing crop yields. The emergence of deep neural network models in the realm of computer vision has brought about a groundbreaking transformation in the swift recognition and surveillance of plant diseases. In this study, we introduce an innovative approach for plant disease prediction, leveraging transfer learning techniques to enable accurate and early detection of plant diseases. The proposed leaf disease prediction framework encompasses multiple key stages, including preprocessing, feature extraction, classification, and postprocessing. Notably, our model excels at feature extraction, utilizing the pre-trained convolutional architecture of VGG16 to extract valuable information from leaf images. These features are then fed into a deep neural network model designed for disease prediction. The feature extraction method employed in our approach excels at capturing intricate disease-related details from plant images. When the deep features extracted are utilized for classification using our proposed classifier, the outcome demonstrates a substantial enhancement in disease prediction performance. Based on extensive experimentation with the benchmark plant village dataset, it is evident that our transfer learning-based approach outperforms existing methods, achieving an impressive accuracy rate of 96.56%. Besides accuracy, the proposed approach surpasses F1 \(\_\) $\_$ score and Kappa scores for the benchmark plant village dataset.