Automated Dimensional Inspection of Knotted Nets Using YOLOv11 and K-Nearest Neighbors Regression

Ensuring consistent mesh geometry is essential for mechanical reliability and quality control of knotted nets in factory production line inspections. This paper presents a fully automated visual inspection system that integrates knot detection, mesh topology reconstruction, and real-world dimensional measurements within a distortion-aware deep learning framework. The knots were detected using a YOLOv11-based object detector trained to localize small and densely arranged targets under varying illumination, deformation, and partial occlusion. To enable accurate metric measurements over large net areas, the images captured using a 150° ultrawide-angle lens were rectified through camera calibration and distortion correction. Knot centroids were extracted from the detected bounding boxes and organized using a row-wise directional linking strategy that reconstructed a diamond-shaped mesh topology while avoiding physically implausible connections. Interknot distances were computed in the rectified image plane and converted into real-world measurements using a K-nearest neighbors (KNN) regression model, which compensated for the residual nonlinearities remaining after distortion correction. The scope of inspection in this study was limited to mesh size and knot spacing evaluation. However, missing knots, broken strands, or other structural defects may affect the dimensional consistency of the net and may be indicated indirectly through abnormal mesh measurements. The proposed system is not intended for direct detection or classification of such defects. It was evaluated on a custom dataset of 1,054 industrial fishing net images, achieving a high detection performance with an mAP50 of 0.99 and mAP50–95 of 0.86. The calibration stage achieved distance estimation errors within ±3.53% relative to the nominal mesh size. The results demonstrated that the proposed approach enabled the automated and quantitative assessment of mesh uniformity, providing a practical and scalable solution for industrial net quality inspection.