Automatic classification of orange fruit diseases using deep neural network model

The classification of orange diseases is essential for reducing financial losses in agriculture and increasing the profitability of orange farming. With the rising global demand for oranges due to their nutritional benefits, maintaining the health of these fruits is critical for public health and economic stability. However, the widespread occurrence of diseases in orange crops heavily impacts productivity and worsens the financial struggles of farmers. To tackle this problem, this study introduces a deep convolutional neural network (CNN) model designed to identify common orange diseases accurately. To improve model robustness, an optimized preprocessing pipeline has been integrated that combines normalization, data augmentation, and noise reduction via Gaussian blur and median filtering. Additionally, a reliable color segmentation technique is presented to increase classification accuracy by extracting highly discriminative features from input images. The dataset, obtained from Kaggle, contains 1790 images across four disease categories: greening, canker, black spot, and fresh. After preprocessing and segmentation, both pre-trained and custom CNN architectures were tested, with the custom, scratch-built CNN achieving the best results. The proposed model reached an accuracy of 97%, showing its effectiveness and potential for use in automated disease detection and crop management systems.