Research On Image Target Recognition Of Millimeter Wave Imaging

At present,the public security problems is highly valued by all countries.And the safety checks of people in public places are also becoming more and more important.The metal security doors,X-ray detectors and other security devices are not comprehensive,low speed and unsafe.Therefore,millimeter wave security equipments which work fast with safety,reliablility and privacy protection have been widely used.Meanwhile,image recognition technolgy,which is one of the impotant components,has also been widely attention.Based on the previous studies,this paper analyzes and studies the target recognition and classification of the millimeter-wave reconstructed images.In respect of millimeter wave image preprocessing,at first,the method of grey-scale stretching is used for image enhancement to highlight the target and reduce the brightness of the impurities.Then,morphological transform is employed to remove impurities and method of OTSU for image binarization.Target would be extracted from the whole image using connected component analysis.Then the feature extraction is performed on the extracted images.The characteristics should be independent of position,scale and orientation and have differences on different targets.Six features are used to describe the characteristics of each target including two low order Hu moment invariants,eccentricity,flat degree,width-length ratio of external rectangle and rectangle degree.In this thesis,300 groups of features are extracted as the basis of image target recognition.One-to-one algorithm based on support vector machine(SVM)is used train the extracted features and recognize dangerous goods,the correct rate reached 93% and comply with the requirement of real-time.At the same time,BP neural network is used for identification and classification of dangerous goods,the recognition rate reached 99.5%,but used more time than support vector machine.Finally,the connected component analysis method is adopted to fine the location of the dangerous goods on the millimeter wave reconstruction images.Experiments show that the the preprocessing methods millimeter-wave reconstructed image are feasible and the pretreatment effect is good.the features extracted in this paper have the property of translation,rotation and scaling invariance.The recognition algorithms has good classification effect and good real-time performance.