基于改进YOLOv8的蓝莓果实成熟度检测方法研究

Research on the Detection Method of Blueberry Fruit Maturity Based on Improved YOLOv8

摘要: 蓝莓是果园种植户经济收益较高的水果之一，识别不同成熟度的蓝莓果实具有重要经济意义，可帮助种植户规划施药方案、估算产量并高效开展采收作业。为提高基于深度学习的蓝莓果实成熟度检测算法的精确性与鲁棒性，提出了一种基于改进YOLOv8的蓝莓果实成熟度检测算法YOLO-BLUE。为了减少参数量，改善目标检测效果，采用BoTNet (Bottleneck Transformer Network)骨干架构代替原YOLOv8的Backbone网络，并引入CoordAttention注意力机制，提升模型对空间结构的理解能力。为了更有效地提高目标检测的准确度，引入边框损失函数WIoU。为验证YOLO-BLUE改进算法的有效性，采集四个成熟度的蓝莓果实并构成数据集，试验结果表明，YOLO-BLUE的精确度、召回率以及平均精度均值达到83.9%、81.6%和87.5%，分别提升4.3%、5.0%和3.7%，为蓝莓果实的成熟度检测算法提供了新的改进思路。

Abstract: Blueberries are one of the fruits with relatively high economic benefits for orchard growers. Identifying blueberry fruits at different ripening stages holds significant economic value, as it can help growers plan pesticide application schemes, estimate yields, and carry out efficient harvesting operations. To enhance the accuracy and robustness of deep learning-based blueberry ripeness detection algorithms, this study proposes an improved blueberry ripeness detection algorithm (YOLO-BLUE) based on YOLOv8. To reduce the number of parameters and improve target detection performance, the backbone architecture of the original YOLOv8 was replaced with BoTNet (Bottleneck Transformer Network). Additionally, the CoordAttention (Coordinate Attention) mechanism was introduced to enhance the model’s ability to understand spatial structures. For more effective improvement of target detection accuracy, the bounding box loss function WIoU was adopted. To verify the effectiveness of the improved YOLO-BLUE algorithm, blueberry fruits at four ripening stages were collected to construct a dataset. Experimental results show that the precision (P), recall (R), and mean average precision at IoU = 0.5 (mAP@0.5) of YOLO-BLUE reached 83.9%, 81.6%, and 87.5%, representing increases of 4.3%, 5.0%, and 3.7% respectively. This study provides a new improvement approach for blueberry ripeness detection algorithms.