A HYBRID DEEP LEARNING FRAMEWORK FOR FRUIT DISEASE DETECTION AND CLASSIFICATION

Abstract

Fruit diseases significantly affect agricultural productivity, quality, and profitability, posing a major challenge to global food systems. Early and accurate disease detection is essential to reduce crop loss and improve yield management. Manual inspection methods require expert knowledge and are often unavailable in rural areas, leading to delayed intervention and inefficient disease control. Conventional image processing and machine learning approaches rely heavily on handcrafted features, making them sensitive to illumination variations, background noise, and complex disease patterns. Additionally, class imbalance, limited labeled datasets, and overlapping visual symptoms further degrade classification accuracy. Traditional systems lack robustness and fail to generalize across multiple fruit varieties and disease conditions. This research proposes an automated fruit disease detection framework using deep learning techniques. A curated image dataset of diseased and healthy fruit samples is pre-processed using normalization, resizing, and data augmentation. Segmentation and enhancement operations are applied to improve lesion visibility. The dataset is then divided into training and testing subsets to ensure reliable model evaluation. Multiple convolutional neural network architectures are implemented, including a custom CNN model and transfer learning-based networks such as VGG16 and VGG19. Optimization strategies such as adaptive learning rate adjustment and dropout regularization are utilized for improved generalization. Model performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. Experimental analysis demonstrates that deep learning-based approaches significantly outperform traditional classifiers. Transfer learning models achieve classification accuracy above 98.27%. The models exhibit strong robustness under variations in background, lighting, and disease severity levels. The findings confirm that deep learning models are effective for fruit disease classification and scalable for real-world agricultural deployment. This study contributes to enhancing intelligent disease diagnosis systems and supports precision agriculture initiatives through automated detection and improved decision-making.