Research on cucumber detection and localization based on Yolo-v5

The cucumber target detection and localization method is one of the core technologies of the cucumber picking robot. However, most of the existing cucumber detection algorithms have problems such as insufficient accuracy and low realtime performance. Based on this practical problem, we propose a lightweight cucumber target detection and localization method based on Yolo-v5 and RGB-D camera for picking robots in order to automatically identify cucumbers in greenhouse. Firstly, proposed an improved Yolo-v5 algorithm for cucumber detection and the experimental results show that in this study, the use of the improved network model can effectively detect cucumbers and improve the identification problem of detecting dense cucumbers. Specifically, the recognition recall, accuracy, mAP and Fl were 95.42%, 85.43%, 93.35% and 67.89%, respectively. Finally, based on detection results in the image and its depth value obtained by Realsense D435i depth camera, the relative distances are calculated for all coordinates which means cucumber localization is achieved. The method can provide technical support for cucumber picking robots to accurately detect and locate cucumber targets in real time.