Research on Apple Detection Algorithm for Picking Robot Based on Improved YOLOv8
Abstract：To address the issues of difficulty in detecting apples under occlusion and the tendency for false detection and missed detection in existing methods, this paper proposes an improved YOLOv8-based apple detection algorithm. The foreground of the apple dataset is enhanced using branches and leaves to strengthen occluded features. A multi-feature information fusion module, C2f-MFIC, is adopted to combine independent detailed information, enriching the content of feature maps and enhancing feature response. A deformable attention mechanism, DAttention, is introduced to enhance the region of interest while suppressing redundant environmental interference. To improve the accuracy of bounding box prediction, Shape-IoU is introduced, considering the shape and size of the position box to strengthen regression positioning. Manhattan distance weighting is used for post-processing of images, combining confidence scores and bounding box thresholds for comprehensive evaluation. Ablation experiments were conducted on the apple detection dataset, and the results show that the improved YOLOv8n algorithm achieves a 3.4% increase in mAP@0.5, a 6.2% increase in mAP@[0.5∶0.95], and a 4.1% increase in recall compared to the base algorithm, with a 0.73% reduction in parameter count and a 3.7% reduction in computational cost. Compared to current popular recognition networks, the proposed method shows better recognition performance for apples.
Keywords：Apple pickingObject detectionYOLOv8MFICDeformable attention
Publication Date：2025-02-27
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：5( 14-18 )