Methods for Mango and Ripeness Detection Using Improved YOLOv8

Automated mango harvesting is a trending practice, yet recognizing mangoes and their ripeness remains a bottleneck due to uneven lighting, fruit overlap, and the similar colors of fruit and leaves. To address these challenges, this paper proposes a method for mango and ripeness detection based on an improved You Only Look Once version 8 (YOLOv8). This method incorporates attention mechanisms and lightweight enhancements to achieve accurate and rapid detection of mangoes and their ripeness under varying lighting conditions, different levels of occlusion, and backgrounds where the colors of fruits and leaves are similar. Validation on 2,742 samples demonstrates that the proposed method based on improved YOLOv8 achieves a mean Average Precision (mAP) of 96.6%, processing each sample in just 22.07 ms. Compared to YOLOv8, the proposed method improves mAP by 1.8% while reducing processing speed by 8.6%. This method provides a precise and rapid detection technology for the automated harvesting of mangoes and potentially other fruits.