A Non-Data-Driven Lightweight Algorithm for Fabric Concave Point Detection via Dual Pointers and Line Fitting

This paper proposes a non-data-driven lightweight fabric concave point detection algorithm based on dual pointers and line fitting: convert images from BGR to HSV color space, filter and sort pixels of specific colors, identify candidate feature points by tracking the increasing or decreasing trend of x-coordinates via dual pointers, locate concave points by fitting lines with the least squares method, and finally filter out invalid points. Both time and space complexity are O(n). Experiments show that the average detection accuracy of 80 test images reaches 0.6933, with a maximum of 0.7977 and a minimum of 0.4999. Errors mainly stem from coordinate recognition deviations and the impact of concave point positions. This algorithm avoids deep learning's reliance on massive data, compensates for the shortcomings of traditional methods, and provides an efficient solution for the intelligent upgrading of the textile industry.