Fruit Ripeness Detector for Automatic Fruit Classification Systems

The development of Artificial Intelligence has led to remarkable advancement in agriculture. Many automatic tools have been developed to reduce human labor and improve accuracy. One of the most popular applications in harvesting and packaging agricultural products is the fruit classification system based on ripeness level. This paper focuses on improving the YOLOv8n architecture by replacing the original convolution operations with a new convolution module called the Receptive Field Convolution Block Attention Module for fruit ripeness detection. This module leverages the advantages of group convolution and Convolution Block Attention Module mechanisms to enhance the feature extraction ability. The experiments are trained and evaluated on the Fruit Ripening Process and Mango And Banana datasets. As a result, the proposed network achieves the best performance at $99.4 \%$ of mAP@0.5 and demonstrates superiority over other methods under the same experimental conditions.