基于机器学习的抛撒地雷夜视智能识别研究

Research on night vision intelligent recognition method of scatterable landmines based on machine learning

利用基于机器学习的抛撒地雷智能探测方法及探测系统,在夜间复杂场景下对不同类型的地雷目标进行智能探测实验研究。在论述抛撒地雷夜视智能识别方法的基础上,以72式防坦克金属地雷、69式防坦克塑壳地雷、58式防步兵橡胶地雷作为抛撒地雷目标,设计不同的实验场景进行被动式夜视智能探测实验,并研究多雷区的探测方法及可行性。结果显示,在给定的实验条件下,抛撒地雷的裸露面积由无明显遮挡、轻微遮挡至部分遮挡的检测精确率分别为98.97%、98.5%、87.5%,召回率分别为99.22%、71.3%、56.8%;在夜间对布设的多雷区进行探测,可有效排除孤立雷并实现对多雷区的标定。表明研究的探雷方法能实现对不同场景下抛撒地雷的智能识别,可用于复杂雷区夜视智能探测技术的进一步研究。

Using machine learning-based intelligent detection methods and systems for scatterable landmines,intelligent detection experiments on different types of landmine targets in complex night scenes are carried out.On the basis of discussing the intelligent identification method for the night vision of scatterable landmines,the 72 type anti-tank metal mines,the 69 type anti-tank plastic shell mines,and the 58 type anti-infantry rubber mines are used as the targets of the scatterable landmines,and different experimental scenes are designed for passive night vision intelligence.The minefield detection method and its feasibility are studied.The results show that under the given experimental conditions,the detection accuracy rates of the exposed area of landmines from no obvious occlusion,slight occlusion to partial occlusion are 98.97%,98.5%,87.5%and the recall rates are 99.22%,71.3%,56.8%.Detecting the deployed minefield at night can effectively eliminate isolated mines and realize the calibration of the minefield.It shows that the mine detection method studied can effectively realize the intelligent detection of the scatterable mines in different scenarios,and can be used for further research of the night vision intelligent detection of complex minefields.