L1-Norm Distance Discriminant Analysis with Multiple Adaptive Graphs and Sample Reconstruction

Linear discriminant analysis (LDA) is sensitive to noise and its performance may decline greatly. Recursive discriminative subspace learning method with an L1-norm distance constraint (RDSL) formulates LDA with the maximum margin criterion and becomes robust to noise by applying L1-norm and slack variables. However, the method only considers inter-class separation and intra-class compactness and ignores the intra-class manifold structure and the global structure of data. In this paper, we present L1-norm distance discriminant analysis with multiple adaptive graphs and sample reconstruction (L1-DDA) to deal with the problem. We use multiple adaptive graphs to preserve intra-class manifold structure and simultaneously apply the sample reconstruction technique to preserve the global structure of data. Moreover, we use an alternating iterative technique to obtain projection vectors. Experimental results on three real databases demonstrate that our method obtains better classification performance than RDSL.

Rolling Bearing Fault Diagnosis Based on Sensitive Feature Transfer Learning and Local Maximum Margin Criterion under Variable Working Condition

In real industrial scenarios, the working conditions of bearings are variable, and it is therefore difficult for data-driven diagnosis methods based on conventional machine-learning techniques to guarantee the desirable performance of diagnosis models, as the models assume that the distributions of both the training and testing data are the same. To enhance the performance of the fault diagnosis of bearings under different working conditions, a novel diagnosis framework inspired by feature extraction, transfer learning (TL), and feature dimensionality reduction is proposed in this work, and dual-tree complex wavelet packet transform (DTCWPT) is used for signal processing. Additionally, transferable sensitive feature selection by ReliefF and the sum of mean deviation (TSFSR) is proposed to reduce the redundant information of the original feature set, to select sensitive features for fault diagnosis, and to reduce the difference between the marginal distributions of the training and testing feature sets. Furthermore, a modified feature reduction method, the local maximum margin criterion (LMMC), is proposed to acquire low-dimensional mapping for high-dimensional feature spaces. Finally, bearing vibration signals collected from two test rigs are analyzed to demonstrate the adaptability, effectiveness, and practicability of the proposed diagnosis framework. The experimental results show that the proposed method can achieve high diagnosis accuracy and has significant potential benefits in industrial applications.

Scatter Matrix Based Domain Adaptation for Bi-Temporal Polarimetric SAR Images

Time series analysis (TSA) based on multi-temporal polarimetric synthetic aperture radar (PolSAR) images can deeply mine the scattering characteristics of objects in different stages and improve the interpretation effect, or help to extract the range of surface changes. However, as far as classification is concerned, it is difficult to directly generate the classification map for a new temporal image, by the use of conventional TSA or change detection methods. Once some labeled samples exist in historical temporal images, semi-supervised domain adaptation (DA) is able to use historical label information to infer the categories of pixels in the new image, which is a potential solution to the above problem. In this paper, a novel semi-supervised DA algorithm is proposed, which inherits the merits of maximum margin criterion and principal component analysis in the DA learning scenario. Using a kernel mapping function established on the statistical distribution of PolSAR data, the proposed algorithm aims to find an optimal subspace for eliminating domain influence and keeping the key information of bi-temporal images. Experiments on both UAVSAR and Radarsat-2 multi-temporal datasets show that, superior classification results with the average accuracy of about 80% can be obtained by a simple classifier trained with historical labeled samples in the learned low- dimensional subspaces.

A Spectral Unmixing Method by Maximum Margin Criterion and Derivative Weights to Address Spectral Variability in Hyperspectral Imagery

Limited to the low spatial resolution of the hyperspectral imaging sensor, mixed pixels are inevitable in hyperspectral images. Therefore, to obtain the endmembers and corresponding fractions in mixed pixels, hyperspectral unmixing becomes a hot spot in the field of remote sensing. Endmember spectral variability (ESV), which is common in hyperspectral images, affects spectral unmixing accuracy. This paper proposes a spectral unmixing method based on maximum margin criterion and derivative weights (MDWSU) to reduce the effect of ESV on spectral unmixing. Firstly, in the MDWSU model, an effective and fast algorithm is employed for establishing the endmember spectral library. Then a spectral weighting matrix based on the maximum margin criterion is constructed based on the endmember spectral library. Besides, derivative analysis and local neighborhood weights are merged into local neighborhood derivative weights, which act as a regularization term to penalize different abundance vectors. Local neighborhood derivative weights and spectral weighting matrix are proved to reduce the effect of ESV. Real hyperspectral data experiments show that the MDWSU model can obtain more accurate endmembers and abundance estimation. In addition, the experimental results, including the spectral angle distance and the root mean square error, prove the superiority of the MDWSU model over the previous methods.