Research on The Identification Technology of Seedless White Grape Stalk Under Small Trellises

In order to achieve accurate identification of grape stems in the natural environment, this study took seedless white grapes in Turpan City, Xinjiang Uygur Autonomous Region as the research object, and proposed an improved grape stem recognition model based on the YOLOv5 model. In order to ensure the recognition effect of fruit stems, Space-to-depth Convolution (SPD-Conv) for low-resolution images and small objects was introduced firstly, and then combined with SE (Squeeze). -and-Excitation) coordinate attention mechanism, and finally the Normalized Wasserstein Distance (NWD) loss function is used. The mAP value of seedless white grape pedicles was 85.3% in the improved recognition model. Compared with YO LOv3-tiny, YOLOv5, YOLOv6, YOLOv8 and other algorithms, the average accuracy is increased by 29.8%, 6.4%, 14.5%, 16.1% and 6.7%.