关键词: 油茶果, 目标检测, YOLOv5算法, 数据增强, 坐标注意力模块

Abstract: In order to solve the problem of low detection accuracy of Camellia oleifera fruit in complex environment, a YOLOv5-CP Camellia oleifera fruit detection method was proposed. Firstly, the RealSense D435i depth camera was used to collect the images of Camellia oleifera fruit in various environments in natural scenes, and the Labelimg software was used to label the Camellia oleifera fruit. Then the cutout data enhancement method and coordinate attention module were introduced, and an improved PANet feature extraction layer was proposed to optimize the YOLOv5 model, and a new detection model YOLOv5-CP of Camellia oleifera fruit was constructed. Finally, YOLOv5-CP was compared with the existing model in the detection of Camellia oleifera fruit in complex environment, the experiments showed that the detection accuracy, recall and average accuracy of YOLOv5-CP model were 98%, 94.6% and 98.4%,  respectively. The detection accuracy of YOLOv5-CP model in occluded and overlapping environments was improved by 11.3% and 108% respectively compared with the original YOLOv5 model. This method effectively improves the detection accuracy of Camellia oleifera fruit in complex environments such as occlusion and overlap, and provides a theoretical basis for the subsequent development of Camellia oleifera picking robot.

Key words: Camellia oleifera fruit, target detection, YOLOv5 algorithm, data enhancement, coordinate attention module