BMPC-YOLOv8: A Lightweight Selenicereus Megalanthus Maturity Detection Alorithm for Natural Picking Scenarios

With the rapid growth in the cultivation area of Selenicereus megalanthus, traditional manual harvesting is inefficient and costly, making automated harvesting a necessary approach. Maturity identification faces challenges such as irregular fruit growth, occlusion and overlap, and non-uniform yellow-green gradient colors.To address this, this paper proposes BMPC-YOLOv8, a lightweight, multi-module collaborative algorithm for maturity detection.The algorithm integrates an improved BAM_ECA mechanism with the efficient HGNetV2 backbone, enhancing feature extraction efficiency and the capability for multi-scale information retention.It introduces the MCA module to enhance the comprehensive understanding of fruit contours, colors, and occlusion relationships. PConv is employed to replace traditional convolution, improving the perception of directional gradients and fine-grained features. Furthermore, it integrates the CBAM module to optimize deep feature maps, reducing confusion between the fruit and the background.Through various experimental validations, BMPC-YOLOv8 demonstrates the best performance in accuracy, speed, and lightweighting. Its mAP@50 increased to 96.5%, model size was reduced by 26.15%, inference speed was boosted by 40.8%, and precision and recall improved by 2.9% and 2.7%, respectively. This research provides an efficient and feasible technical solution for the automated harvesting and quality grading of Selenicereus megalanthus, possessing significant practical application value.