A deep learning approach to detect diseases in pomegranate fruits via hybrid optimal attention capsule network

Abstract

In 2022, the production rate of pomegranate is estimated at approximately 4.8 million metric tons. Unfortunately, these fruits are susceptible to many different kinds of diseases caused by bacterial, viral, and fungal infections. Such diseases can have a major negative impact on fruit quality, production, and the profitability of pomegranate cultivation. Nowadays, several machine learning and deep learning methods are used to identify pomegranate fruit diseases automatically and effectively. In post-harvest pomegranate fruit disease detection, deep learning has great potential to extract complex patterns and features from large datasets. This can improve disease identification accuracy, enabling more efficient disease control, lower crop losses, and better resource management. The proposed work introduces an intelligent deep learning-based approach for accurately detecting pomegranate diseases, begins with Improved Guided Image Filtering (Improved GIF) and resizing to pre-process fruit images, followed by feature extraction (shape, color, texture) using GLCM and GLRLM to streamline classification. Extracted features are then fed into a novel Hybrid Optimal Attention Capsule Network (Hybrid OACapsNet), which classifies the images as normal or diseased, conditions such as bacterial blight, heart rot, and scab. Our analysis indicates that the proposed classifier has a classification accuracy of 99.19 %, precision of 98.45 %, recall of 98.41 %, F1-score of 98.43 %, and specificity of 99.45 % compared to other techniques. So this approach offers a framework, which is a feasible solution for automated detection of diseases in fruits, thereby benefiting farmers and supporting their farming operations.