An Optimal Hyper Parameter Selection of Deep Learning Models for Automatic Citrus Fruit and Leaves Diseases Detection

This study focuses on enhancing the detection of citrus fruit and leaf diseases to reduce financial losses and improve product quality in the agricultural sector. Citrus diseases, including black spot, canker, scab, greening, and melanosis, lead to significant rejections of otherwise healthy fruits. To address this, we propose an Improved Convolutional Neural Network (ICNN) optimized with the Wild Horse Optimization (WHO) algorithm, which overcomes the challenges of manual CNN hyper parameter tuning. The WHO-ICNN model extracts discriminative features from multiple layers to enhance disease classification accuracy. Experimental comparisons with other models demonstrate that WHO-ICNN performs better detecting citrus diseases. These findings suggest that the proposed model can provide farmers with a more efficient and reliable tool for citrus disease management, ultimately improving yield quality and profitability.