Ground surface subsidence is a universal phenomenon in coal mining areas which can cause serious damage to the surrounding environment. In this paper, we consider the use of differential interferometric synthetic aperture radar (D-InSAR), multi-temporal InSAR (MT-InSAR), and the pixel offset tracking technique to monitor the surface deformation of a coal mining area. In this study, we use the two-pass D-InSAR method to generate 19 interferometric image pairs from 20 TerraSAR-X SpotLight images. The results show that D-InSAR can be used to obtain high accuracy surface deformation in the mining areas where there is no high gradient deformation, and the pixel offset tracking method offers advantages in those areas where high gradient deformation is found, but its performance is not stable. This means that the unilateral use of these technologies cannot obtain reliable subsidence information in mining areas. Therefore, it is essential to find a new way to integrate the respective advantages of these different methods. In this paper, a new fusion method combining the D-InSAR result with the offset tracking result based on a spatial decorrelation distribution map is proposed to obtain the subsidence results in a mining area. To ensure the reliability of the results, a decision rule is proposed for the spatial decorrelation distribution map, which is generated manually by union analysis in ArcGIS. In the experiments, the mean absolute error of the fusion result is 0.0748 m, while that of D-InSAR is 0.1890 m, and that of offset tracking is 0.1358 m. It is therefore clear that the proposed fusion method is more reliable and more accurate than the use of individual methods, and it may be able to serve as a reference in mining subsidence monitoring.

This paper introduces a novel semi-supervised tri-training classification algorithm based on regularized local discriminant embedding (RLDE) for hyperspectral imagery. In this algorithm, the RLDE method is used for optimal feature information extraction, to solve the problems of singular values and over-fitting, which are the main problems in the local discriminant embedding (LDE) and local Fisher discriminant analysis (LFDA) methods. An active learning method is then used to select the most useful and informative samples from the candidate set. In the experiments undertaken in this study, the three base classifiers were multinomial logistic regression (MLR), k-nearest neighbor (KNN), and random forest (RF). To confirm the effectiveness of the proposed RLDE method, experiments were conducted on two real hyperspectral datasets (Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and Reflective Optics System Imaging Spectrometer (ROSIS)), and the proposed RLDE tri-training algorithm was compared with its counterparts of tri-training alone, LDE, and LFDA. The experiments confirmed that the proposed approach can effectively improve the classification accuracy for hyperspectral imagery.