YOLO-TP: A process detection model of tunnel real scene based on gradient information interaction

• Published in: 2024 5th International Conference on Computer, Big Data and Artificial Intelligence (ICCBD+AI) pp. 530 - 536
• Main Authors: Fan, Jiaying, Pan, Shuping, Jiang, Xinbo, Zhang, Shengtao, Liu, Hongliang, Wang, Chuan, Li, Jianghua, Ke, Chenglin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2024
• Subjects: Accuracy; Artificial intelligence; Benchmark testing; Big Data; Computational modeling; Data models; Dataset; Feature extraction; Image resolution; Object detection; Process Recognition; Safety; Tunnel Detection
• DOI: 10.1109/ICCBD-AI65562.2024.00094

Target detection technology is an effective method to detect tunnel operation safety. Although the field has made some progress in detection models, the lack of datasets and detection models has hindered its development. We produce and propose Tunnel Processes data (TP), a challenging dataset of high-precision real-world tunnel process target detection that reduces the limitations of research in this field. TP has a high resolution: 1920×1080, 1734images,9 categories. TP is the first tunnel process data set under construction to date, and its fidelity exceeds that of other tunnel process datasets in the field of target detection. In addition, we also propose a YOLO-TP benchmark model to make up for the lack of a high-precision tunnel process image target detection baseline model. Experiments on TP datasets demonstrate their validity, and experiments on the dataset presented in this paper demonstrate the validity of the benchmark model.