A Novel Method for Concealed Target Detection and Classification Based on Passive Millimeter Wave Imaging

Passive millimeter wave(PMMW) imaging, which possesses advantageous features as no radiation, noncontact to human body, concealed targets detecting, plays an important role when applied to public security check. The traditional PMMW image classification methods rely on features extracted by experts manually. However, it is more difficult to extract exact features manually, given the low resolution and less information of the PMMW images. Deep learning (DL) based on convolutional neural network (CNN) has the characteristics of automatic feature extraction. In this paper, a target classification method based on CNN is proposed to classify the PMMW images. Better effects are realized by the proposed CNN model which combines the structure of LeNet model and the structure construction idea of VGGNet, with the advantages of uncomplicated structure and few parameters. The problem of low recognition accuracy of PMMW image in fewer samples cases is also solved. The measured data shows that the overall classification accuracy of the proposed algorithm is 98.17% on the testing set images. For a single category of guns, knives and other small objects, the accuracy is 100%, 95.8% and 98.7%, respectively. The measured data also indicates that compared with traditional algorithms of Machine Learning and other DL method, under the same image samples condition, the effects of proposed algorithm is obviously improved.