YOLOv8-CML: A lightweight target detection method for Color-changing melon ripening in intelligent agriculture

Abstract

Abstract To enhance the efficiency of robot automatic picking of Color-changing melons under intelligent agriculture environments, this study introduces a lightweight model for target detection, YOLOv8-CML, for effectively detecting the ripeness of Color-changing melons. The model structure is simplified to reduce the deployment cost of image recognition models on agricultural edge devices. First, we replace the Bottleneck structure of the C2f module with a Faster Block, which reduces superfluous computations and the frequency of memory accesses by the model. Then, we use a lightweight C2f module combined with EMA attention in Backbone, which can efficiently collect multi-scale spatial information and reduce the interference of background factors on Color-changing melon recognition. Next, we use the idea of shared parameters to redesign the detection head to perform the Conv operation uniformly before performing the classification and localization tasks separately, thus simplifying the structure of the model. Finally, we use the α-IoU approach to optimize the CIoU loss function, which can better measure the overlap between the predicted and actual frames to improve the accuracy of the recognition. The experimental results show that the parameters and FLOPs ratio of the improved YOLOv8-CML model decreased by 42.9% and 51.8%, respectively, compared to the YOLOv8n model. In addition, the model size is merely 3.7MB, and the inference speed is increased by 6.9%, along with mAP@0.5, Precision, and FPS. Our proposed model provides a vital reference for deploying Color-changing melon picking robots.