Research On Classification Of Unexploded Ordnance Based On Transient Electromagnetic Detection

Due to the advantages of high sensitivity and strong anti-interference ability,the transient electromagnetic method(TEM)is widely used in the detection of underground metal targets such as unexploded bombs.However,this method has a sensitive response to harmful unexploded bombs and harmless metal shrapnel,so the false positive rate of detection is high,resulting in extremely time-consuming follow-up cleaning work.Earlier studies have shown that unexploded bombs and unexploded harmful targets have certain differences in characteristic response attenuation,but no characteristic value with high enough discrimination ability can be found to classify them.Aiming at this problem,this paper detects unexploded bombs and harmless targets based on the portable system and the vehicle-mounted system,uses the characteristic responses of different draws of the target body and the fitting parameters to construct the feature value,and uses the support vector machine(SVM)and K-Nearest Neighbors(KNN)classification algorithm based on the data is used to build a classification model,and finally achieve accurate classification.Firstly,the basic principle of the transient electromagnetic method is introduced,the dipole model of the unexploded bomb is given and the characteristic response of the unexploded bomb is analyzed,and the sensors and system structures of the portable detection system and the vehicle-mounted detection system used in the experiment are introduced.Then,the algorithm principle of SVM and KNN is given,the concept and the formula basis of support vector and hyperplane are introduced;the feature value selection and classification are introduced in detail to build the model,select the training data set,and call the classification algorithm to train the classification model according to different feature values.Finally,in the experimental part,twelve typical types of unexploded bombs and eighteen kinds of harmless targets were selected,and the targets were buried at different distances and depths in the field.According to the detection data of the portable system and the vehicle-mounted system,two types of different feature values and train the classification model,and then perform testing and noise analysis.Experimental results show that the classification models with normalized characteristic response and characteristic response parameters as feature value can accurately divide the classification sample set.The accuracy of the classification models constructed by SVM and KNN algorithms both reached more than 95%.Comparing the two construction methods of feature value,we can see that classification based on characteristic response has the advantages of fast speed and simple processing.Large,noise and channel selection are easily affected by noise interference,resulting in noise samples in the data set,which has a significant impact on classification accuracy.In contrast,in the classification based on fitting parameters,noise has less influence on the parameters,the feature value are more stable,and it has higher accuracy and robustness.