The non-negative matrix factorization (NMF) algorithm represents the original image as a linear combination of a set of basis images. This image representation method is in line with the idea of "parts constitute a whole" in human thinking. The existing deep NMF performs deep factorization on the coefficient matrix. In these methods, the basis images used to represent the original image is essentially obtained by factorizing the original images once. To extract features reflecting the deep localization characteristics of images, a novel deep NMF architecture based on underlying basis images learning is proposed for the first time. The architecture learns the underlying basis images by deep factorization on the basis images matrix. The deep factorization architecture proposed in this paper has strong interpretability. To implement this architecture, this paper proposes a deep non-negative basis matrix factorization algorithm to obtain the underlying basis images. Then, the objective function is established with an added regularization term, which directly constrains the basis images matrix to obtain the basis images with good local characteristics, and a regularized deep non-negative basis matrix factorization algorithm is proposed. The regularized deep nonlinear non-negative basis matrix factorization algorithm is also proposed to handle pattern recognition tasks with complex data. This paper also theoretically proves the convergence of the algorithm. Finally, the experimental results show that the deep NMF architecture based on the underlying basis images learning proposed in this paper can obtain better recognition performance than the other state-of-the-art methods.

The estimation of the sea background from maritime infrared video sequences is necessary for target detection. It is challenging to obtain an accurate background model when the scene is a complex and fluctuating sea surface. However, the amplitude spectrum sequences at each frequency point of the pure seawater frames in the Fourier domain are more stable than the gray value sequences of each pixel in the spatial domain. Thus, the background is modeled with a Gaussian distribution in the Fourier domain with the mean and variance adapted over time. In addition, the sea background dynamics are introduced by the variance of the amplitude spectrum sequence, and the dynamic Gaussian discriminant coefficient is set up for each frequency point. Two target discrimination flags are presented based on the background dynamics and Gaussian discriminant process to extract the target more accurately. Furthermore, the entropy filter in the Fourier domain is designed to enhance the target and suppress the sea clutter. The proposed method is successfully tested over several maritime infrared video sequences and has better detection effects compared with several existing state-of-the-art methods.

This paper presents an incremental learning algorithm for the hybrid RBF-BP (ILRBF-BP) network classifier. A potential function is introduced to the training sample space in space mapping stage, and an incremental learning method for the construction of RBF hidden neurons is proposed. The proposed method can incrementally generate RBF hidden neurons and effectively estimate the center and number of RBF hidden neurons by determining the density of different regions in the training sample space. A hybrid RBF-BP network architecture is designed to train the output weights. The output of the original RBF hidden layer is processed and connected with a multilayer perceptron (MLP) network; then, a back propagation (BP) algorithm is used to update the MLP weights. The RBF hidden neurons are used for nonlinear kernel mapping and the BP network is then used for nonlinear classification, which improves classification performance further. The ILRBF-BP algorithm is compared with other algorithms in artificial data sets and UCI data sets, and the experiments demonstrate the superiority of the proposed algorithm.

Remote sensing (RS) multi-spectral images are usually suffered from cloud and fog cover, which canlead to analysis troubles and application limitations. An ovel patch-based dark channel prior dehaze method is proposed for solving this problem. An Atmospheric light (AL) curved surface hypothesis, instead of globally invariable plane, is applied to describe AL distribution, and apatch-based approach is given to estimate curved surface. By using AL curved surface estimation, a new recovering model for RS multi-spectral images is given to obtain dehaze-free images. Comparative experiments are conducted, those results illustrate that the proposed method can produces visually impressive restored images, and the proposed method is superior to other relative methods interms of image quality evaluations.

This paper presents a novel magneto-optical isolator based on an ultra-wideband and high efficiency photonic crystals (PCs) waveguide and gyromagnetic ferrites. The three-dimensional numerical simulation finds that the photonic crystals waveguide’s (PCW) transmission efficiency rises with its height and width. The corresponding experiments are performed by using a triangular lattice Al₂O₃ dielectric posts array in 5G millimeter wave band. The measured transmission efficiency is up to 90.78% for the optimal PCs waveguide structure, which has ultra-wide operating bandwidth from 23.45 to 31.25 GHz. The magneto-optical isolator is designed by inserting two rectangular gyromagnetic ferrites into the PCs waveguide. Due to the contrast between the effective permeability of the left and right circular polarization waves passing through the magnetized ferrite sheets, the ferromagnetic resonance absorption of the forward and reverse waves is different. By using finite element method, the isolation is optimized to be 49.49 dB for the isolator and its relative bandwidth reaches 8.85%. The high isolation, broadband, and easy integration indicate that our designed magneto-optical isolator has significant advantage in 5G communication systems.