Enhancing Rubus idaeus L. maturity detection and classification: A YOLO-CG-CPCA network model for efficient feature extraction and precise analysis

Accurate and efficient identification of Rubus idaeus L. (raspberry) ripeness remains a challenging task due to subtle maturity variations and complex natural environments. This study focuses on improving the accuracy and efficiency of Rubus idaeus L. ripeness classification by developing an improved YOLOv8 model. A standardized dark-box imaging system was developed and employed based on mobile imaging to ensure high consistency and stability in image acquisition. To address fine-grained ripeness differences, a C2f-ContextGuided module was incorporated. It enhances both local and global feature extraction, improving feature discrimination. Additionally, a Channel Prior Convolutional Attention mechanism was introduced to highlight key information in the feature maps. The Wise-IoUv2 loss function was also integrated to improve localization robustness. Experimental results indicate that the proposed YOLO-CG-CPCA model outperformed the baseline YOLOv8s by 3.0 % in mean Average Precision. Moreover, it achieved relative improvements of 1.8 %, 0.9 %, 0.9 %, 2.5 %, and 1.5 % over Faster R-CNN, YOLOv5s, YOLOv7-Tiny, YOLOv10s, and YOLOv11s, respectively. In conclusion, the proposed model significantly enhances the accuracy and efficiency of Rubus idaeus L. ripeness identification. It offers technical support for intelligent post-harvest grading and provides a reference for the automated detection of other small fruits and vegetables, such as strawberries and blueberries.