scholarly journals Nonnegative Signal Decomposition with Supervision

2013 ◽  
Vol 2013 ◽  
pp. 1-8
Author(s):  
Teng Li ◽  
Huan Chang ◽  
Jun Wu
Keyword(s):  
Objective Function ◽  
Image Classification ◽  
Pattern Classification ◽  
Optimization Problem ◽  
Real Data ◽  
Classification Performance ◽  
Signal Decomposition ◽  
Weak Supervision ◽  
Training Samples ◽  
Novel Algorithm

This paper presents a novel algorithm to numerically decompose mixed signals in a collaborative way, given supervision of the labels that each signal contains. The decomposition is formulated as an optimization problem incorporating nonnegative constraint. A nonnegative data factorization solution is presented to yield the decomposed results. It is shown that the optimization is efficient and decreases the objective function monotonically. Such a decomposition algorithm can be applied on multilabel training samples for pattern classification. The real-data experimental results show that the proposed algorithm can significantly facilitate the multilabel image classification performance with weak supervision.

scholarly journals A DIVERSIFIED DEEP BELIEF NETWORK FOR HYPERSPECTRAL IMAGE CLASSIFICATION

2016 ◽  
Vol XLI-B7 ◽  
pp. 443-449
Author(s):  
P. Zhong ◽  
Z. Q. Gong ◽  
C. Schönlieb
Keyword(s):  
Image Classification ◽  
Real World ◽  
Hyperspectral Image ◽  
Classification Performance ◽  
Fine Tuning ◽  
Hyperspectral Image Classification ◽  
Latent Factors ◽  
Deep Model ◽  
Training Samples ◽  
Limited Training Samples

In recent years, researches in remote sensing demonstrated that deep architectures with multiple layers can potentially extract abstract and invariant features for better hyperspectral image classification. Since the usual real-world hyperspectral image classification task cannot provide enough training samples for a supervised deep model, such as convolutional neural networks (CNNs), this work turns to investigate the deep belief networks (DBNs), which allow unsupervised training. The DBN trained over limited training samples usually has many “dead” (never responding) or “potential over-tolerant” (always responding) latent factors (neurons), which decrease the DBN’s description ability and thus finally decrease the hyperspectral image classification performance. This work proposes a new diversified DBN through introducing a diversity promoting prior over the latent factors during the DBN pre-training and fine-tuning procedures. The diversity promoting prior in the training procedures will encourage the latent factors to be uncorrelated, such that each latent factor focuses on modelling unique information, and all factors will be summed up to capture a large proportion of information and thus increase description ability and classification performance of the diversified DBNs. The proposed method was evaluated over the well-known real-world hyperspectral image dataset. The experiments demonstrate that the diversified DBNs can obtain much better results than original DBNs and comparable or even better performances compared with other recent hyperspectral image classification methods.

scholarly journals PolSAR Image Classification Using a Superpixel-Based Composite Kernel and Elastic Net

2021 ◽  
Vol 13 (3) ◽  
pp. 380
Author(s):  
Yice Cao ◽  
Yan Wu ◽  
Ming Li ◽  
Wenkai Liang ◽  
Peng Zhang
Keyword(s):  
Image Classification ◽  
Feature Fusion ◽  
Classification Performance ◽  
Elastic Net ◽  
Synthetic Aperture ◽  
Polarimetric Synthetic Aperture Radar ◽  
Training Samples ◽  
Composite Kernel ◽  
Limited Training Samples ◽  
State Of Art

The presence of speckles and the absence of discriminative features make it difficult for the pixel-level polarimetric synthetic aperture radar (PolSAR) image classification to achieve more accurate and coherent interpretation results, especially in the case of limited available training samples. To this end, this paper presents a composite kernel-based elastic net classifier (CK-ENC) for better PolSAR image classification. First, based on superpixel segmentation of different scales, three types of features are extracted to consider more discriminative information, thereby effectively suppressing the interference of speckles and achieving better target contour preservation. Then, a composite kernel (CK) is constructed to map these features and effectively implement feature fusion under the kernel framework. The CK exploits the correlation and diversity between different features to improve the representation and discrimination capabilities of features. Finally, an ENC integrated with CK (CK-ENC) is proposed to achieve better PolSAR image classification performance with limited training samples. Experimental results on airborne and spaceborne PolSAR datasets demonstrate that the proposed CK-ENC can achieve better visual coherence and yield higher classification accuracies than other state-of-art methods, especially in the case of limited training samples.

scholarly journals A DIVERSIFIED DEEP BELIEF NETWORK FOR HYPERSPECTRAL IMAGE CLASSIFICATION

2016 ◽  
Vol XLI-B7 ◽  
pp. 443-449 ◽  
Author(s):  
P. Zhong ◽  
Z. Q. Gong ◽  
C. Schönlieb
Keyword(s):  
Image Classification ◽  
Real World ◽  
Hyperspectral Image ◽  
Classification Performance ◽  
Fine Tuning ◽  
Hyperspectral Image Classification ◽  
Latent Factors ◽  
Deep Model ◽  
Training Samples ◽  
Limited Training Samples

In recent years, researches in remote sensing demonstrated that deep architectures with multiple layers can potentially extract abstract and invariant features for better hyperspectral image classification. Since the usual real-world hyperspectral image classification task cannot provide enough training samples for a supervised deep model, such as convolutional neural networks (CNNs), this work turns to investigate the deep belief networks (DBNs), which allow unsupervised training. The DBN trained over limited training samples usually has many “dead” (never responding) or “potential over-tolerant” (always responding) latent factors (neurons), which decrease the DBN’s description ability and thus finally decrease the hyperspectral image classification performance. This work proposes a new diversified DBN through introducing a diversity promoting prior over the latent factors during the DBN pre-training and fine-tuning procedures. The diversity promoting prior in the training procedures will encourage the latent factors to be uncorrelated, such that each latent factor focuses on modelling unique information, and all factors will be summed up to capture a large proportion of information and thus increase description ability and classification performance of the diversified DBNs. The proposed method was evaluated over the well-known real-world hyperspectral image dataset. The experiments demonstrate that the diversified DBNs can obtain much better results than original DBNs and comparable or even better performances compared with other recent hyperspectral image classification methods.

scholarly journals Label consistent transform learning for pattern classification

2019 ◽  
Vol 13 ◽  
pp. 174830261988139
Author(s):  
He-Feng Yin ◽  
Xiao-Jun Wu
Keyword(s):  
Image Processing ◽  
Pattern Classification ◽  
Dictionary Learning ◽  
Optimization Problem ◽  
Classification Performance ◽  
Alternative Strategy ◽  
Learning Approaches ◽  
Training Time ◽  
Practical Applications ◽  
Consistency Constraint

Transform learning has been successfully applied to various image processing tasks in recent years. Nevertheless, transform learning learns the representation in an unsupervised fashion. To make transform learning suitable for pattern classification, we introduce a label consistency constraint into transform learning and propose a new label consistent transform learning to enhance the classification performance of transform learning. The resulting optimization problem can be solved elegantly by employing the alternative strategy. Experimental results on publicly available databases demonstrate that label consistent transform learning outperforms several dictionary learning approaches and the recently proposed discriminative transform learning. More importantly, label consistent transform learning has the least training time which has the potential in practical applications.

Exergy and sensibility analysis of each individual effect in a kraft multiple effect evaporator

TAPPI Journal ◽  
2019 ◽  
Vol 18 (10) ◽  
pp. 607-618
Author(s):  
JÉSSICA MOREIRA ◽  
BRUNO LACERDA DE OLIVEIRA CAMPOS ◽  
ESLY FERREIRA DA COSTA JUNIOR ◽  
ANDRÉA OLIVEIRA SOUZA DA COSTA
Keyword(s):  
Optimization Problem ◽  
Real Data ◽  
Kraft Pulping ◽  
Steam Temperature ◽  
Input Flow ◽  
Transfer Coefficients ◽  
Heat Transfer Coefficients ◽  
Multiple Effect ◽  
Sensibility Analysis ◽  
Exergetic Analysis

The multiple effect evaporator (MEE) is an energy intensive step in the kraft pulping process. The exergetic analysis can be useful for locating irreversibilities in the process and pointing out which equipment is less efficient, and it could also be the object of optimization studies. In the present work, each evaporator of a real kraft system has been individually described using mass balance and thermodynamics principles (the first and the second laws). Real data from a kraft MEE were collected from a Brazilian plant and were used for the estimation of heat transfer coefficients in a nonlinear optimization problem, as well as for the validation of the model. An exergetic analysis was made for each effect individually, which resulted in effects 1A and 1B being the least efficient, and therefore having the greatest potential for improvement. A sensibility analysis was also performed, showing that steam temperature and liquor input flow rate are sensible parameters.

Shape Optimization of Hydraulic Structures: an Example of an Optimum Design of a Fish Passage

10.29007/2k64 ◽  
2018 ◽  
Author(s):  
Pat Prodanovic ◽  
Cedric Goeury ◽  
Fabrice Zaoui ◽  
Riadh Ata ◽  
Jacques Fontaine ◽  
...  
Keyword(s):  
Numerical Model ◽  
Shape Optimization ◽  
Objective Function ◽  
Optimum Design ◽  
Optimization Problem ◽  
A Priori ◽  
Coupled System ◽  
Fish Passage ◽  
Hydraulic Structures ◽  
Design Variables

This paper presents a practical methodology developed for shape optimization studies of hydraulic structures using environmental numerical modelling codes. The methodology starts by defining the optimization problem and identifying relevant problem constraints. Design variables in shape optimization studies are configuration of structures (such as length or spacing of groins, orientation and layout of breakwaters, etc.) whose optimal orientation is not known a priori. The optimization problem is solved numerically by coupling an optimization algorithm to a numerical model. The coupled system is able to define, test and evaluate a multitude of new shapes, which are internally generated and then simulated using a numerical model. The developed methodology is tested using an example of an optimum design of a fish passage, where the design variables are the length and the position of slots. In this paper an objective function is defined where a target is specified and the numerical optimizer is asked to retrieve the target solution. Such a definition of the objective function is used to validate the developed tool chain. This work uses the numerical model TELEMAC- 2Dfrom the TELEMAC-MASCARET suite of numerical solvers for the solution of shallow water equations, coupled with various numerical optimization algorithms available in the literature.

scholarly journals Semi-Supervised PolSAR Image Classification Based on Self-Training and Superpixels

2019 ◽  
Vol 11 (16) ◽  
pp. 1933 ◽  
Author(s):  
Yangyang Li ◽  
Ruoting Xing ◽  
Licheng Jiao ◽  
Yanqiao Chen ◽  
Yingte Chai ◽  
...  
Keyword(s):  
Image Classification ◽  
State Of The Art ◽  
Sample Selection ◽  
Spatial Relations ◽  
Speckle Noise ◽  
Classification Performance ◽  
Selection Strategy ◽  
Training Set ◽  
Polarimetric Synthetic Aperture Radar ◽  
Unlabeled Sample

Polarimetric synthetic aperture radar (PolSAR) image classification is a recent technology with great practical value in the field of remote sensing. However, due to the time-consuming and labor-intensive data collection, there are few labeled datasets available. Furthermore, most available state-of-the-art classification methods heavily suffer from the speckle noise. To solve these problems, in this paper, a novel semi-supervised algorithm based on self-training and superpixels is proposed. First, the Pauli-RGB image is over-segmented into superpixels to obtain a large number of homogeneous areas. Then, features that can mitigate the effects of the speckle noise are obtained using spatial weighting in the same superpixel. Next, the training set is expanded iteratively utilizing a semi-supervised unlabeled sample selection strategy that elaborately makes use of spatial relations provided by superpixels. In addition, a stacked sparse auto-encoder is self-trained using the expanded training set to obtain classification results. Experiments on two typical PolSAR datasets verified its capability of suppressing the speckle noise and showed excellent classification performance with limited labeled data.

scholarly journals A simulated annealing-based algorithm for selecting balanced samples

2021 ◽  
Author(s):  
Roberto Benedetti ◽  
Maria Michela Dickson ◽  
Giuseppe Espa ◽  
Francesco Pantalone ◽  
Federica Piersimoni
Keyword(s):  
Simulated Annealing ◽  
Optimization Problem ◽  
Sample Selection ◽  
Auxiliary Information ◽  
Real Data ◽  
Simulation Experiments ◽  
Balanced Sampling ◽  
Inclusion Probabilities ◽  
Random Method ◽  
Annealing Algorithms

AbstractBalanced sampling is a random method for sample selection, the use of which is preferable when auxiliary information is available for all units of a population. However, implementing balanced sampling can be a challenging task, and this is due in part to the computational efforts required and the necessity to respect balancing constraints and inclusion probabilities. In the present paper, a new algorithm for selecting balanced samples is proposed. This method is inspired by simulated annealing algorithms, as a balanced sample selection can be interpreted as an optimization problem. A set of simulation experiments and an example using real data shows the efficiency and the accuracy of the proposed algorithm.

Sign in / Sign up

Export Citation Format

Share Document