A Deep Learning Method for Recognizing Types of Unexploded Ordnance Based on Magnetic Detection
Zhu Wen, Songtong Han, Chengwei Gao, Yuze Chen, Limei Guo, Ya Zhang
Abstract
The concealment of unexploded ordnance (UXO) left behind by wars and live firing exercises is a problem, not only posing a serious threat to the safety of local residents, but also bringing great difficulties to explosive disposal work. The magnetic detection of UXO has the advantages of portability and efficiency, but it is difficult to recognize the types of UXO through magnetic moment estimation. A deep learning method for recognizing types of UXO in response to the current difficulties in magnetic detection is proposed. By designing a magnetic flux gate array acquisition system and conducting magnetic detection experiments on UXO simulated targets, the effective detection distance of this method is found to be about 2.2 m. The accuracy of recognizing three types of UXO simulated targets is greater than 95.8% and the F1-score is larger than 92.5%. The accuracy is higher than 85.9% and the F1-score is greater than 81.1% under the effects of interference. This method can suppress the influences of environmental magnetic fields, providing a technical reference for recognizing types of UXO based on magnetic detection.