Swiftlet Nest Arch Height Measurement System Using YOLO11n-seg

Abstract

This research develops an automatic system for measuring the arch height of swiftlet nests based on computer vision with the YOLO11n-seg model. This system overcomes the limitations of manual measurement at PT. Waleta Asia Jaya, where the arch height is one of the characteristics used to determine the quality of the product. The dataset contains 300 images captured using ESP32-S3 CAM (resolution 640×480). Images in the dataset is preprocessed using red channel extraction, intensity-based cropping, background correction, and morphology processing. The nest contour is analyzed using polar transformation and peak detection. The evaluation includes four approaches such as No Smoothing Filter, Ramer-Douglas-Peucker (RDP) Filter, Median Filter (MedFilt), and Savitzky-Golay (SavGol) Filter. The No Smoothing approach provides the best accuracy with a Root Mean Square Error (RMSE) of 2.28 millimeters and a coefficient of determination of 0.7753, as well as a time saving of 78.26%. ElasticNet analysis shows Lpred as the main predictor (0.4994) and vertical height as a secondary contributor (0.4614). This system improves the efficiency of quality control with an error below 3 millimeters.