High-resolution Remote Sensing Recognition Based On Sparse Constraint Restricted Boltzmann Machine

With the continuous development and progress of the sensor technology, the spatial resolution of the remote sensing image is continuously improved, and the target feature information in the high resolution image is more abundant. Originally used for low resolution remote sensing image feature extraction and target detection method has been unable to meet the application of high resolution remote sensing image, the high resolution remote sensing images automatically or semi automatically classification and recognition is facing unprecedented challenges. In recent years, the rapid development of deep learning provides an effective method to solve the problem.The sparse coding method is adopted in the human visual information system, which can easily identify the complex object in the high resolution remote sensing image. Based on Lorentz sparse constraints restricted Boltzmann machine(RBM) of high resolution remote sensing image target recognition method is studied in this paper. It based on the depth of learning theory, simulation of the human visual system information processing mechanism, sparse constraint is introduced into the restricted Boltzmann machine, of the remote sensing image in the feature information of the target is extracted effectively, realize the reliability of the target recognition of high resolution remote sensing image. The main contents of the study include:(1) A Cauchy Lorentz RBM sparse constraint(function RBM LRBM) model based on Lorentz distribution is proposed in this paper. Based on this model, the performance of the target feature extraction is evaluated, and the results are also analyzed.(2) A deep network based on Lorentz sparse constraint RBM is constructed, and the performance of the depth network is analyzed by experiments.(3) The model of combining LRBM with SVM is proposed, which improves the reliability of the target recognition of high resolution remote sensing image. First, LRBM as high resolution remote sensing image feature extractor, extracting feature information of the target and effective, removes the redundant information. Then, using SVM to achieve the target recognition. Experiments and Analysis on two kinds of target recognition in remote sensing images have been carried out. It is proved that the combination of SVM and LRBM is an effective method for target recognition in high resolution remote sensing image.(4) The influence of the depth of the network model on the reliability of image target recognition is also studied in this paper, the experiment proves that the single layer LRBM can meet the requirements of the high resolution remote sensing image feature extraction.