Disease detection and classification in agricultural products has always been a headache for farmers because they directly affect the productivity and quality of products. Scientists have found and proposed many different research directions to solve this problem, and computer vision technology has received special attention and research recently. In this paper, we propose a new method based on the improved YOLOv8 model, integrated with the Feature Pyramid Network (FPN) to improve disease classification ability. The proposed network architecture uses Efficient Net as the backbone for feature extraction, combined with FPN before the detection head to enhance multi-scale representation. Experiments were conducted on the “Fruit and Vegetable Disease - Healthy and Rotten" dataset, resulting in a top-1 accuracy of 97.81%, 0.4% better than the original YOLOv8 model. These findings highlight the effectiveness of our architectural modifications in improving the classification performance of fruit diseases.