Work

Capsicum Weight and Size Estimation Using YOLO11-Segmentation Model with Machine Learning Integration

Year: 2026
Type: conference-paper
Abstract: This research investigates the application of a custom-trained YOLO11-seg segmentation model to estimate the size and weight of capsicum fruits. A dataset of 350 greenhouse images, captured at optimal distances, was utilized for training the model in Google Colaboratory, ensuring efficient processing. The YOLO11-seg model demonstrated strong performance, attaining a mean Average Precision (mAP) of 0.798 at an Intersection over Union (IoU) threshold of 0.5, along with an overall segmentation accuracy of 81% for capsicum instances. On the test dataset, the maximum confidence score for capsicum segmentation reached 91%, with an average confidence of$83\% \pm 5.4\%$. The YOLO11s-seg model exhibited a compact footprint (19.7 MB) and efficient performance, achieving an average inference time of 26 ms per image. Correlation analysis indicated strong positive relationships between segmented mask area and weight (0.88), as well as between mask perimeter and both length (0.83) and diameter (0.92). Additionally, the mask area and perimeter were highly correlated (0.98). Linear regression models developed for predicting weight and size from segmentation features achieved R2values of 0.87 and 0.93, respectively. In addition, non-linear approaches such as Support Vector Regression (SVR) and Random Forest were evaluated, with the Random Forest model outperforming the others, delivering R2values of 0.94 for weight prediction and 0.98 for both length and diameter estimation. The average prediction errors were recorded as 17.37 g for weight, 12.56 mm for length, and 8.40 mm for diameter. Overall, these findings demonstrate the reliability and practical applicability of the YOLO11-seg model in agricultural environments.
Language: English
Subfield: Plant Science
Open Access status: hybrid
OpenAlex ID: W7202293232
FWCI:
Cites: 15
Cited by:
Related to: 10