Image Noise Level Estimation by Neural Networks

This study aims to develop new procedures in optimal neuro fuzzy modeling for time series data. Specifically in this research, the development of new procedure in modeling fuzzy Takagi-Sugeno-Kang order one for time series data which parameter-parametemic determination is done by singular value and neural network decomposition method, in order to obtain method of forming neuro fuzzy model for time series data optimal. In this research, we have developed a procedure to get the optimal Takagi-Sugeno-Kang fuzzy model for time series data by optimizing the parameter value search in consequence of fuzzy rule using singular value decomposition method. A new model of neuro fuzzy modeling is optimized, the fuzzy model whose parametem optimization is based on the neural network by the singular value decomposition method. Parameters in consequent part of the rule of fuzzy are optimized by the singular value decomposition method and the parameters in the antecedent part of the fuzzy rule are optimized based on neural network backpropagation with gradient descent method.

5G is the fifth generation of wireless communication technologies for cellular networks. Massive Multiple In and Multiple Out (MIMO) is one of the technologies being applied in 5G wireless communication systems. Massive MIMO uses large number of antennas to increase sector throughput and capacity without using additional spectrum resources. As the number of antennas increases, the computational complexity grows dramatically, and this involves matrix calculations with complex numbers. Due to the computational complexity, traditional general purpose processors cannot achieve real-time and high-throughput processing in matrix data processing. To accelerate the process, many researchers have proposed hardware designs for specific matrix operations. However, MIMO related matrix processing requires multiple matrix operations and single-function matrix processors cannot achieve the best acceleration. Therefore, a universal matrix arithmetic processor is required for modern wireless communication systems. This dissertation introduces a general matrix arithmetic processor to accelerate these calculations including matrix multiplication, Singular Value Decomposition (SVD), and QR decomposition for MIMO systems in wireless communications. For MIMO receivers, linear decoding and channel estimation involve calculations of matrix multiplication and matrix inverse. Non-linear decoding (sphere decoding) requires QR decompositions to form a triangular matrix for searching. For MIMO transmitters, power assignment, channel rank, and channel quality estimation need SVD to calculate the singular values. Channel estimation also requires matrix multiplication and inverse. As SVD and QR decomposition are both numerically stable methods for calculating a matrix inverse, we find matrix multiplication, SVD, and QR decomposition are fundamental and commonly used matrix operations for MIMO systems. The proposed design unifies matrix multiplication, SVD and QR decomposition into one systolic array to best utilize hardware resources and reduce data transmission time. The design can be implemented in fixed-point or single-precision floating-point depending on the requirements of the application. The system behavior can be controlled by instructions such as elementary multiplication, rotation and vector projection, which allows the system to work as a coprocessor in a baseband System on a Chip (SoC). The proposed design can be used in the Software Defined Radio (SDR) community to speed up the processing time and development period. Our system is designed in Simulink, which allows wireless communications designers can verify their algorithms and map them to hardware. The fixed-point design can be implemented in scenarios when the RF front end is fixed. The floating-point design is a general design; it can easily process data from different kinds of resources or can be used for real-time analysis. The design is implemented and verified in FPGA hardware on a Xilinx Zynq UltraScale+ ZCU111 Evaluation Platform. The design is implemented in matrix sizes of 88 and 1616 using fixed-point and single precision floating-point numbers. We achieve one to two orders of magnitude performance improvement for SVD over a software implementation.

This paper proposes a simple and accurate estimation of the additive white Gaussian noise for the noise-contaminated digital images. One can easily estimate the noise level through singular value decomposition (SVD) to the noise-polluted image if an image is deteriorated by the additive white Gaussian noise. As described in the paper, the sum of some specific singular values has the linear relationship with the standard deviation of noise. Based on no correlation between noises, we add known noises upon a noise image. Then noise level is estimated by solving a nonlinear over-determined matrix equation. The proposed algorithm was experimentally tested by the benchmark images and outperforms estimation method of selecting weak textured patches using principal component analysis (PCA). The proposed method is more independent on the original image information and presents a higher accuracy and a stronger robustness for a range of noise level in various images.

This paper introduces the design method of face recognition system is a kind of BP neural network. This design uses the forward backward propagation algorithm of multi-layer neural network, BP algorithm, BP neural network for face recognition was built by using MATLAB software, and the use of facial feature a large number of data for training the neural network, the neural network can distinguish different features. The experimental results show that using this method for face recognition has a good effect.

Through to analyze the characteristics of the digital cartoon image, proposed a based on discrete cosine transform (DCT) and singular value decomposition (SVD) combined contourlet transform with zero watermarking algorithm, the algorithm first for image contourlet transform and then make the low-frequency subband DCT and SVD decomposition again after operation block, the largest singular value as the image feature extracting for zero watermarking structure. In order to improve the security of the algorithm, introducing into visual cryptography processing the watermark image, the zero watermarking features obtained and secret sharing for the logic operation is zero watermarking characteristics of the final value.. The simulation results show that the algorithm has strong robustness.

It is well known that neural networks (NN) with backpropagation (BP) are used for recognition and learning. The basic networks have three layers, input layer, one hidden layer and output layer, and the scale a of 3-layered NN depends on the number of hidden layer units (fixed number of input and output layer units on NN). In this paper the authors make a multi (4,5)-layered NN with four or five layers on BP (input layer, two or three hidden layers, output layer) and try to compare a 3-layered NN and a multi-layered NN, in terms of the convergence. As a result, the convergence of a multilayered NN is very low compared with a 3-layered NN. However, a multilayered NN extracts two meanings from learning data such as shape and density,.

The number of eigenmodes in an optical fibre depends on the wavelength of the excitation laser beam as well as on the exact geometry and refractive index profile of the fibre. The latter is often proprietary information and, when available, is only specified to within manufacturing tolerances. We here present a method for obtaining the number of fibre modes, as well as their shape, which requires no knowledge about the fibre, save a very approximate core radius. The method is based on the singular value decomposition (SVD) of a set of speckle patterns, measured at the output end of the fiber, which is then expanded onto a set of orthonormal basis functions. We present two possible approaches for the field expansion, where the first approach uses a generic orthonormal basis, such as Laguerre–Gaussian or Zernike functions, and the second one is a basis-free approach where the set of speckled patterns themselves form the basis. Using a set of simulated speckles patterns, we observed that the correct number of fibre modes can be obtained through the SVD decomposition, even at high levels of additive random noise. With a slight extension, using speckle patterns obtained at multiple excitation wavelengths (or equivalently, for different lengths of the same fiber) the method can also retrieve the shape of the actual fibre modes, by forming an appropriate linear combination of SVD modes.

This study innovatively introduces the Singular Value Decomposition (SVD) method to deeply analyze the heat transfer characteristics of energy piles, aiming to address the issue of insufficient attention to the complexity of heat transfer within piles in existing research. The results show that the heat transfer within the pile, due to the involvement of different materials and changes in heat sources, has a slower rate of singular value decline, exhibiting higher mode diversity. In contrast, the heat transfer between piles has a faster trend of singular value decline, with relatively simple and stable main heat transfer modes. This study not only verifies the effectiveness of the SVD method in analyzing the heat transfer characteristics of energy piles, but also provides a new perspective for the design and optimization of energy piles. It reveals the significant differences in heat transfer characteristics between within-pile and between-pile, providing a theoretical basis for improving the efficiency and stability of geothermal exchange systems.

In this present work, a technique for discrimination between normal and cirrhotic liver segmented regions of interest (SROIs) based on singular value decomposition (SVD) of GLCM matrix is reported. Thirty four B-mode ultrasound images taken from 22 normal volunteers and 12 patients suffering from liver cirrhosis were collected from Department of Radio diagnosis and Imaging, PGIMER, Chandigarh, India. Firstly, the gray level co-occurrence matrix (GLCM) texture features are computed for 121 SROIs (82 normal SROIs, 39 cirrhotic SROIs) and classification is done using a neural network (NN) classifier. The classification accuracy of 95.86% is achieved without feature selection. Secondly, feature selection is carried out by two different approaches. In approach 1, standard correlation based feature selection (CFS) is used to find the optimal subset of GLCM texture features which provides best discrimination between normal and cirrhotic SROIs. It has been observed that CFS method,results in an optimal subset of 7 GLCM texture features {angular second moment (ASM), Contrast, Variance, Sum Average, Entropy, Difference Entropy and Information Measures of Correlation-1}. In approach 2, the potential of singular values obtained by singular value decomposition (SVD) of GLCMs for discrimination between normal and cirrhotic SROIs is investigated. It has been observed that only first 2 singular values can provide effective discrimination between normal and cirrhotic liver SROIs. In the classification stage a neural network (NN) classifier is used. The classification accuracy of 95.04% is obtained in both cases. From the comparison it is concluded that only first two singular values obtained by SVD decomposition of the GLCMs and a NN classifier can be used to build acomputationally efficient computer aided diagnostic (CAD) system for predicting liver cirrhosis.

The singular value decomposition is a matrix decomposition technique widely used in the analysis of multivariate data, such as complex space-time images obtained in both physical and biological systems. In this paper, we examine the distribution of singular values of low-rank matrices corrupted by additive noise. Past studies have been limited to uniform uncorrelated noise. Using diagrammatic and saddle point integration techniques, we extend these results to heterogeneous and correlated noise sources. We also provide perturbative estimates of error bars on the reconstructed low-rank matrix obtained by truncating a singular value decomposition.

This paper presents a new approach to evolutionary artificial neural networks, based on the integration of feedforward neural networks, messy genetic algorithms (GAs), and singular value decomposition (SVD). The set of competing hidden nodes with variable number of connections from the input layer represents an evolving neural network. Selection of hidden nodes is based on their estimation via SVD. The resulting singular values are used to determine significance of hidden nodes for the network's output. To represent connectivity of hidden nodes and to process the topology of connections between input and hidden layers, we employ the approach of messy GAs. This establishes a framework for processing strings of variable length which codes this topology and allows one to search for useful combinations of input variables. The proposed approach is tested using sonar data classification.

Aiming at an additive white noise, a method to estimate the standard deviation of white noise is presented in the paper. The sampled data are first transformed to the wavelet domain by wavelet transform with the different scale levels and the different lengths of the compactly supported wavelets, and then the wavelet coefficients at each level are employed to estimate the standard deviation of the noise directly, which avoids the sequencing operation of the wavelet coefficients carried out in the method to estimate the standard deviation by a median value. Examples prove that the method presented in the paper has the advantage over the method to estimate the standard deviation by a median value. In addition, it can be seen from the results calculated at the different scale levels and different lengths of the compactly supported wavelets that the scale level has a bigger influence on the estimation of the standard deviation of the noise, and that the length of the compactly supported wavelets has only a maller influence on the estimation of the standard deviation of the noise.

We Propose PSO-BP network traffic prediction algorithm which based on BP neural network and improved by the particle swarm optimization. Use PSO algorithm to optimize the initial weight and threshold values of BP network, and use history to train BP neural network and realize the simulation by MATLAB. The results show that, PSO-BP algorithm can improve network traffic prediction accuracy and speed up the convergence rate of BP network.

A new approach based on multiwavelets transformation and singular value decomposition (SVD) is proposed for the classification of image textures. Lower singular values are truncated based on its energy distribution to classify the textures in the presence of additive white Gaussian noise (AWGN). The proposed approach extracts features such as energy, entropy, local homogeneity and max-min ratio from the selected singular values of multiwavelets transformation coefficients of image textures. The classification was carried out using probabilistic neural network (PNN). Performance of the proposed approach was compared with conventional wavelet domain gray level co-occurrence matrix (GLCM) based features, diserete multiwavelets transformation energy based approach, and HMM based approach. Experimental results showed the superiority of the proposed algorithms when compared with existing algorithms.

Investigating two types of seismic signals often misclassified in practice, seismic events and quarry blasts, two novel approaches using singular value decomposition (SVD) to extract features of the main wavelet coefficients (WCs) and product function components (PFs) and the support vector machine (SVM) to classify them are proposed. This research collected and preprocessed 200 seismic events and 200 quarry blasts from the Yongshaba mine, China. Discrete wavelet transform (DWT) and local mean decomposition (LMD) were used to decompose the signals into several WCs and PFs, respectively, and the correlation coefficient and variance contribution ratio were used to select the main WCs and PFs. Finally, the singular value features of the selected six WCs and PFs, which can discriminate between seismic events and quarry blasts, were extracted, and the features were input to backpropagation (BP) neural network, Bayes, SVM, and logistic regression (LR) classifiers. The results show that SVD can effectively extract signal features, and that the SVM classifier offers better classification results than the BP neural network, Bayes, and LR classifiers. In addition, the LMD-SVD-SVM-based method is better than the DWT-SVD-SVM-based method in accuracy, specificity, and sensitivity, with values of 96.0 %, 97.0 %, and 95.0 %, and 95.5 %, 97.0 %, and 94.0 %, respectively. Therefore, DWT and LMD based on SVD and SVM techniques provide useful approaches to seismic event and quarry-blast classification.