Year: 2024
Type: article
Abstract: The traditional longitudinal tear detection methods have poor visible light imaging quality, poor generalization ability, and high hardware costs, then an improved encoder-decoder network was proposed to solve the longitudinal tear detection problem. This method is a line structured light for image acquisition system. Firstly, the input images are downscaled by a sorting algorithm; secondly, the input feature map is semantically segmented by the reduced-dimensional encoder-decoder network, and the pixel segmentation result is mapped to the interval of the longitudinal tear; Finally, the positioning and length of the longitudinal tear are calculated after the semantic segmentation result is back-projected to the world coordinate system. The experimental results show that: The DICE and MIOU of the improved network are 97.69% and 95.47%, respectively; The recall and precision of the improved detection are 96.60% and 95.67%, respectively; The maximum relative error of length measurement is 15.04%.
Source: Measurement and Control
Authors: Lei You, Minghua Luo, Xinglin Zhu, Bin Zhou
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,000
Award: G6755878630
OpenAlex ID: W4402541657