Key points are not available for this paper at this time.
Cherry ripeness detection is a critical step in achieving intelligent harvesting. There are three core challenges associated with cherry ripeness detection: (1) complex environmental interference makes it difficult to extract surface features of cherries; (2) the detection of dense small targets results in high rates of missed detections and false positives; (3) the computational load of the model is too high to be deployed on edge agricultural devices. Based on these challenges, this study proposes a lightweight real-time detection model, CMD-YOLO (Cherry Maturity Detection-YOLO), based on an improved YOLOv12 architecture to optimize cherry ripeness detection performance. First, we propose the P23 structure-adaptive detection head with optimized number and scale, selectively removing redundant modules to enhance feature perception for dense small objects. Second, we innovatively propose the CDCHead module, which employs depthwise convolutions and SE attention mechanisms for cascaded channel fusion. This enables refined extraction of multi-scale spatial features while suppressing complex background interference. Finally, we introduce the Shape-IoU loss function. By precisely modeling variations in bounding box shape and scale, it enhances detection accuracy while maintaining low computational complexity. Experiments show that on the Multi-Scenario Cherry Ripeness Dataset V1, compared to the baseline model YOLOv12, the detection accuracy of this model is improved to 70.7 % (an increase of 5.3 %), the recall rate reaches 70.0 % (a rise of 10.3 %), the average precision mAP50 is 74.3 % (a rise of 11.8 %), and mAP50:95 of 54.9 % (an increase of 10.4 %), while reducing the number of model parameters to 0.7 M (a decrease of 73.1 %). The CMD-YOLO detection framework proposed in this study performs exceptionally well in detecting dense, small cherry targets in complex environments, with overall performance superior to that of existing mainstream models.
Li et al. (Tue,) studied this question.