Research On Passive Millimeter Wave Image Target Recognition Algorithm

Passive millimeter wave(PMMW)imaging technology acquires millimeter wave brightness temperature images of target scenes by receiving their millimeter wave radiation energy.An important research direction at present is interferometric synthetic aperture radiometer imaging system.Based on the applications of PMMW detection,this paper establishes a PMMW interferometric synthetic aperture radiometer image set,and applies machine learning theory to target recognition algorithms for PMMW images.The main research work includes:(1)Research on PMMW image target recognition algorithm based on traditional feature extraction methods.In order to suppress noise interference and information loss in the PMMW images,first,local binary patterns(LBP)algorithm and corner detection algorithm are utilized to extract the local textural features and the corner features of the PMMW images,then support vector machine(SVM)is applied to classify the images based on two features respectively.By introducing a fusion coefficient,the fusion recognition framework based on LBP-Corner feature is constructed,to improve recognition performance.The validity of the algorithm is verified by the recognition experiments on the PMMW image dataset.(2)Research on PMMW image target recognition algorithm based on sparse representation methods.First,a super-complete dictionary of PMMW images is constructed.Then,the dictionary is updated to obtain a robust sparse representation based on the image set itself,using different dictionary learning methods.Thus,the drawback of the artificial feature extraction methods is reduced,and the information loss of a single image sample is complemented by the connection between the images,and the influence of noise on the recognition results is suppressed by the robust sparse representation.Through the recognition experiments on the PMMW image dataset,the recognition performance of sparse representation algorithm based on different dictionary learning methods is analyzed.(3)Research on PMMW image target recognition algorithm based on deep learning methods.First,denoising auto-encoder(DAE)is utilized to suppress the noise of the PMMW images,and reconstructed PMMW images are obtained.Then,convolutional neural network(CNN)extracts robust deep features of the reconstructed images for target recognition.Through the migration and fusion of DAE reconstruction network and CNN target recognition network,an end-to-end target recognition algorithm framework based on DAE-CNN is constructed.From separate pre-training of sub-networks to the overall joint optimization of the whole network,a direct mapping from PMMW images to labels is established.The recognition experiments on the PMMW image dataset show that the target recognition algorithm based on deep learning obtains better recognition performance and has good application value.