Arcing detection method of metro pantograph based on improved yolov4-tiny
Arcing on metro pantographs contribute to the accelerated wear of pantograph strips and the catenary, imposing a serious safety risk to rail transit. For the problem of strong light interference and variable background in pantograph arcing detection for metro vehicles, an arcing detection method bas...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Department of Electric Drive for Locomotives
2023-07-01
|
| Series: | 机车电传动 |
| Subjects: | |
| Online Access: | http://edl.csrzic.com/thesisDetails#10.13890/j.issn.1000-128X.2023.04.012 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Arcing on metro pantographs contribute to the accelerated wear of pantograph strips and the catenary, imposing a serious safety risk to rail transit. For the problem of strong light interference and variable background in pantograph arcing detection for metro vehicles, an arcing detection method based on an improved yolov4-tiny model is proposed. In order to improve the detection ability of small targets, this method incorporated a third-scale prediction branch to the original two-scale prediction branches of yolov4-tiny, to enable the positioning of small arcs in the shallow network. Moreover, a receptive field block (RFB) module was embedded beneath the backbone network, to expand the network′s receptive field and enhance the model′s feature extraction ability. The test results show the average precision (<italic>P</italic><sub>AP</sub>) of the improved model on the test set is increased to 98.2%, marking an improvement of 7.8 percentage points over yolov4-tiny. Despite yielding similar positioning effect for arcs, the improved model works at significantly higher speeds than yolov4, enabling inference speed of a single image in only 6.5 ms. The proposed method is proven effective and accurate in fulfilling the arcing detection task for pantographs of metro vehicles. |
|---|---|
| ISSN: | 1000-128X |