Title: Deep learning-based visual detection method for defects in automotive flowing water turn signal lights

Authors: Guoyang Wan; Hanqi LI; Chengwen Wang; Bingyou Liu; Jincheng Chen; Hong Zhang

Addresses: Department of Electrical Engineering, Anhui Polytechnic University, Beijing Road No. 8, QI, 241000, China ' Department of Electrical Engineering, Anhui Polytechnic University, Beijing Road No. 8, QI, 241000, China ' Department of Electrical Engineering, Anhui Polytechnic University, Beijing Road No. 8, QI, 241000, China ' Department of Electrical Engineering, Anhui Polytechnic University, Beijing Road No. 8, QI, 241000, China ' Department of Electrical Engineering, Anhui Polytechnic University, Beijing Road No. 8, QI, 241000, China ' Department of Electrical Engineering, Anhui Polytechnic University, Beijing Road No. 8, QI, 241000, China

Abstract: To address the challenge of high-speed detection of vehicle flowing water turn signal lights, a novel detection method based on deep learning is proposed. This approach aims to achieve swift and efficient detection of flowing water lights. The proposed method incorporates an innovative deep learning detector designed to dynamically measure key points of the flowing water light and its operational status. Furthermore, this paper emphasises the optimisation of the anchor-free object detection network through the utilisation of the heatmap regression method for detecting industrial objects. The incorporation of an attention module and the refinement of the heatmap generation methodology facilitate improvements in both the detector's accuracy and the speed of training process convergence. The outcomes of the experiment demonstrate that the detection system devised in this paper exhibits commendable speed, accuracy, and user-friendly operation.

Keywords: deep learning; defect detection; object detection; visual inspection; turn signal light.

DOI: 10.1504/IJAISC.2025.151127

International Journal of Artificial Intelligence and Soft Computing, 2025 Vol.9 No.3, pp.233 - 250

Received: 05 Sep 2024
Accepted: 03 Sep 2025
Published online: 14 Jan 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article