A Vision-Based Two-Stage Fruit Outer Diameter Detection Method

Regarding the mobile monitoring system in agricultural automation, the outer diameter monitoring of greenhouse vegetables has significant implications for accurately regulating fruit growth and cultivation plans. In this paper, a novel vision-based two-stage outer diameter detection method is proposed, including the segmentation and outer diameter calculation of greenhouse tomato images. To address the issues of multiple tomatoes in a single image and the similarity in color between tomatoes and the background, a YOLOv8 network embedded with convolutional block attention module is constructed for detection and segmentation in the first stage. Additionally, to solve the problem of inaccurate outer diameter detection caused by different tomato poses and pedicles, a defined central axis method is proposed, which can adaptively filter out the maximum outer diameter candidates and calculate accurate values utilizing depth information in the world coordinate system. The backbone network is fine-tuned on the pre-trained YOLOv8 model with 1000 tomato images. Extensive measurement experiments, involving 215 tomatoes in 80 images, is conducted to validate the effectiveness of our proposed approach. Compared to the ground truth, we obtain a measurement error of 4.27 mm, the coefficient of determination value R2 is 0.9349.