Research On Detection And Identification Methods Of Hidden Prohibited Items Based On Passive Millimeter Wave Imaging

Taking automatic detection and recognition of concealed contraband using passive millimeter wave imaging as application background, an in-depth and meticulous research on the detection, feature extraction and recognition of PMMW images is conducted in this paper.The detection of concealed contraband is implemented by pre-processing, including image de-noising, enhancement and segmentation. The wavelet soft threshold de-noising algorithm is improved for image de-noising, the simulation results show that this method allows images to retain relatively sharp edges while the background noise is suppressed. FT visual saliency algorithm is introduced for image enhancement in this paper, and as for the millimeter wave image is single channel image, the original processing method applied in Lab color space is convert to a single channel color space, so the original three-dimensional feature vectors are made into one dimension, which reduce the operation time while guarantee the process effect. It is proved that the concealed contraband can be highlight and maintain the integrity of the shape information no matter their gray value is higher or lower than the human body using this method. Then after segmentation using OTSU algorithm, morphological reconstruction algorithm is added to exclude the background area for extracting the target area.This paper analyzes the features and shortcomings of Fourier operator, invariant moments, the geometrical constant feature extraction methods, and propose a feature extraction method based on the Radon transform. The simulation results show the superiority of this method in operational, the range of expression, simplicity, stability, sensitivity, etc.In recognition part, an algorithm integrating the support vector machine (SVM) classifier with features extracted based on the Radon transform is proposed for automatic identification of concealed contraband by analyzing and comparing the performance of random forests, RPROP neural networks and SVM. Experimental results show that the algorithm can quickly and accurately recognize the concealed contraband.Finally, with VS2008, OPENCV and MFC, an interactive interface is constructed for concealed contraband recognition using the C++ language.