Accurate Apple Fruit Stalk Cutting Technology Based on Improved YOLOv8 with Dual Cameras

Highlights The length of the apple stalk retained at picking affects the quality of the harvested fruit. Improving deep learning networks for accurate detection of apples and apple stalks. Proposing a two-camera, two-stage process for shearing fruit stalks. Modeling and evaluating apple stalk shearing machines to validate the feasibility of shearing methods. Abstract. When robotically picking apples, a stalk that is too short affects freshness, and a stalk that is too long scratches the skin of the apple. This article describes an accurate apple fruit stalk cutting technology based on improved YOLOv8 with dual cameras, which retains the proper stalk length during shearing. The approach incorporates the GAM attention mechanism into YOLOv8 to improve the recognition of apples and their stalks. The proposed method uses a two-stage shearing process: The first stage involves a depth camera at the device’s initial end to recognize and localize apples in 3D, guiding the end actuator towards the nearest fruit. In the second stage, a camera on the end actuator identifies and determines the position of the stalks. Tests showed that the improved YOLOv8-GAM for apple stalk recognition improved the mAP value by 3.7% compared to the original YOLOv8 model. In the experiment of the shearing device, at the initial distance of 500 mm, at the end of the movement, measured by infrared range finder, the systematic error value of the single camera to recognize the distance between the center of the knife head and the fruit stalk was 5.83 mm, while the dual camera was only 3.88 mm. The dual-camera cutting technology for apple stalks can make it possible to retain the proper stalk length when cutting apples, which can reduce the damage of apples and avoid secondary manual shearing, and improve the precision and efficiency of smart apple picking. Keywords: Automatic shearing device, Deep learning, Fruit stalk identification, Intelligent agricultural machines.