Anomaly detection (AD) is an important technique for hyperspectral image processing and analysis. Typically, it is accomplished by extracting knowledge from the background and distinguishing anomalies and background using the difference between them. However, it is almost impossible to obtain “pure” background to achieve an ideal detection because of anomaly contamination. The low-rank and sparse matrix decomposition (LRaSMD) technique has been proved to have the potential to solve the aforementioned problem. But the accuracy and time consumption need to be further improved. Thus we propose a local hyperspectral AD method based on LRaSMD with an optimization algorithm for better performance. The LRaSMD technique is first implemented with semisoft Go decomposition (GoDec) rather than GoDec to quickly and accurately set the background apart from the anomalies. Then the low-rank prior knowledge of the background is fully explored to compute the background statistics. After that, the local Mahalanobis distance of pixels is calculated with the sliding dual-window strategy to detect the probable anomalies. The proposed method is validated using four real hyperspectral data sets with ground-truth information. Our experimental results indicate that the proposed method achieves better detection performance as compared with the comparison algorithms.
Nonnegative matrix factorization (NMF) is a widely used method of hyperspectral unmixing (HU) since it can simultaneously decompose the hyperspectral data matrix into two nonnegative matrices. While traditional NMF cannot guarantee the sparsity of the decomposition results and remain the geometric structure during the decomposition. On the other hand, deep learning, with carefully designed multi-layer structures, has shown great potential in learning data representation and been widely used in many fields. In this paper, we proposed a graph-regularized and sparsityconstrained deep NMF (GSDNMF) for hyperspectral unmixing. The deep NMF structure was acquired by unfolding NMF into multiple layers. To improve the unmixing performance, the L1 regularizers of both the endmember and abundance matrices were used to add sparsity constraint. And the graph regularization term in each layer was also incorporated to remain the geometric structure. Since the model is a multi-factor NMF problem, it is difficult to optimize all the factors simutaneously. In order to acquire better intializations for the model, we proposed a layer-wise pretraining strategy to initialize the deep network based on the efficient NMF solver, NeNMF. An alternative update algorithm was also proposed to further fine-tune the network to obtain the final decompositon results. Experiments on both the synthetic data and real data demonstrate that our algorithm outperforms several state-of-art approaches.
Hyperspectral unmixing (HU) refers to the process of decomposing the hyperspectral image into a set of endmember spectra and the corresponding set of abundance fractions. Non-negative matrix factorization (NMF) has been widely used in HU. However, most NMF-based unmixing methods have single-decomposition structures, which may have poor performance for highly mixed and ill-conditioned data. We proposed a sparsity-constrained multilayer NMF (MLNMF) method for spectral unmixing of highly mixed data. The MLNMF structure was established by decomposing the abundance matrix layer-by-layer to acquire the endmember matrix and the abundance matrix in the next layer. To reduce the space of solutions, sparsity constraints were added to the multilayer model by incorporating an L1 regularizer to the abundance matrix in each layer. Moreover, a layerwise strategy based on the Nesterov’s optimal gradient method was also proposed to optimize the multifactor NMF problem. Experiments on both synthetic data and real data demonstrate that our proposed method outperforms several other state-of-art approaches.
Aircraft roll is an important reason for geometric distortion of airborne push-broom hyperspectral images. A two-step rectification method is proposed to correct the geometric distortions of push-broom hyperspectral images caused by aircraft roll. According to the distortion characteristics of hyperspectral images, image distortion is mainly classified into two categories: high frequency distortion and large amplitude distortion. For high frequency distortion, the first-step correction is implemented by slide matching of segmented windows between adjacent scan lines. For large amplitude distortion, the second-step correction is achieved using ground control points and linear features between a reference map and an uncorrected image. Experimental results show that the proposed method can eliminate image distortion caused by aircraft roll effectively.
The convolution neural network (CNN) is becoming more and more powerful in many areas such as image classification and speech recognition. Some projects begin to apply it on mobile phones, but often need plenty of time due to the huge amount of computation. This paper uses a deep learning framework named MXNet to realize the forward process on the mobile phone. In order to lower the time it costs, we focus on how to use the other computing device on the chip—the mobile GPU. We choose the OpenCL to offload the most time consuming layer in the CNN—convolution layer to the GPU. Besides that, this paper makes several improvements to achieve better performance and finally the experimental results demonstrate that the forward process only takes half the time in our algorithm. To the best of the authors’ knowledge, this work is the first published implantation of CNN accelerating on mobile GPU.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.