关键词: 草莓采摘, 图像处理, 神经网络, 果柄叶识别

Abstract: In order to reduce the damage to strawberry fruit during the picking process and achieve the recognition and localization of strawberry stems in complex practical picking scenarios, this paper proposes a hybrid approach combining improved YOLOv4 and traditional image processing techniques. Unlike conventional methods, this approach places the picking point on the strawberry stem to ensure the integrity of the fruit during picking. Firstly, the improved YOLOv4 algorithm is employed to accurately locate the strawberries. Then, image processing techniques are utilized to segment the fruit, strawberry stems, and background, thereby determining the image coordinates of the picking point. Finally, the spatial coordinates of the picking point are measured by integrating binocular positioning algorithms and distance sensor data. Experimental results demonstrate that the improved YOLOv4 algorithm achieves a testing accuracy of 90% and, in conjunction with traditional image processing techniques, effectively eliminates interference from complex backgrounds, thereby enhancing the robustness of the algorithm. The depth distance error of strawberry stems measured by the distance sensor is within 5mm, enabling precise localization of the picking point, thus exhibiting superior applicability in practical picking scenarios.

Key words: strawberry picking, image processing, neural networks, stem leaf identification