Integrating genetic algorithm method with neural network for land use classification using SZ- 3 CMODIS data

This paper presents a methodology on land use mapping using CMODIS (Chinese Moderate Resolution Imaging Spectroradiometer) data on-board SZ-3 (Shenzhou 3) spacecraft. The integrated method is composed of genetic algorithm (GA) for feature extraction and neural network classifier for land use classification. In the data preprocessing, a moment matching method was adopted to remove the stripes in the images. Then by using the reproduction, crossover and mutation operators of GA based on the mechanism of “natural selection”, and with Jeffries-Matusita distance as its discriminate rule and the training samples, the optimal band combination for land use classification was obtained. To generate a land use map, the three layers back propagation neural network classifier is used for training the samples and classification. Compared with the Maximum Likelihood classification algorithm, the results show that the accuracy of land use classification is obviously improved by using our proposed method, the selected band number in the classification process is reduced, and the computational performance for training and classification is improved. The result also shows that the CMODIS data can be effectively used for land use/land cover classification and change monitoring at regional and global scale.     

Research on remote sensing image of land cover classification based on multiple classifier combination

This paper proposed an algorithm in which the maximum probability and the weighted average strategy were used for the combination of member classifiers. Using parallel computing, we test the algorithm on a China-Brazil Earth Resources Satellite (CBERS) image for land cover classification. The results show that using three computers in parallel can reduce the classification time by 30%, as compared with using only one computer with a dual core processor. The accuracy of the final image is 93.34%, and Kappa is 0.92. Multiple classifier combination can enhance the precision of the image classification, and parallel computing can increase the speed of calculation so that it becomes possible to process remote sensing images with high efficiency and accuracy.     

Polarimetric Synthetic Aperture Radar Image Classification by a Hybrid Method

Different methods proposed so far for accurate classification of land cover types in polarimetric synthetic aperture radar (SAR) image are data specific and no general method is available. A novel hybrid framework for this classification was developed in this work. A set of effective features derived from the coherence matrix of polarimetric SAR data was proposed. Constituents of the feature set are wavelet, texture, and nonlinear features. The proposed feature set has a strong discrimination power. A neural network was used as the classification engine in a unique way. By exploiting the speed of the conjugate gradient method and the convergence rate of the Levenberg-Marquardt method (near the optimal point), an overall speed up of the classification procedure was achieved. Principal component analysis (PCA) was used to shrink the dimension of the feature vector without sacrificing much of the classification accuracy. The proposed approach is compared with the maximum likelihood estimator (MLE) based on the complex Wishart distribution and the results show the superiority of the proposed method, with the average classification accuracy by the proposed method (95.4%) higher than that of the MLE (93.77%). Use of PCA to reduce the dimensionality of the feature vector helps reduce the memory requirements and computational cost, thereby enhancing the speed of the process.     

Land cover classification of remotely sensed image with hierarchical iterative method

Land cover classification is one of the most wide ly used applications of remote sensing. The use ofmultitemporal remote sensing data in land cover clas sification is one of the effective methods of obtainingaccurate land cover/land use data. For a particularimage, different land cover types often show a similarspectral response and are difficult to separate. Theadvantage of using multitemporal data is that differentvegetation types show different spectral characteristicsin…     

Application of multi-scale feature extraction to surface defect classification of hot-rolled steels

Feature extraction is essential to the classification of surface defect images. The defects of hot-rolled steels distribute in different directions. Therefore, the methods of multi-scale geometric analysis (MGA) were employed to decompose the image into several directional subbands at several scales. Then, the statistical features of each subband were calculated to produce a high-dimensional feature vector, which was reduced to a lower-dimensional vector by graph embedding algorithms. Finally, support vector machine (SVM) was used for defect classification. The multi-scale feature extraction method was implemented via curvelet transform and kernel locality preserving projections (KLPP). Experiment results show that the proposed method is effective for classifying the surface defects of hot-rolled steels and the total classification rate is up to 97.33%.     

Universal Steganalysis of Data Hiding in Grayscale Images

This paper proposed an universal steganalysis program based on quantification attack which can detect several kinds of data hiding algorithms for grayscale images. In practice, most techniques produce stego images that are perceptually identical to the cover images but exhibit statistical irregularities that distinguish them from cover images. Attacking the suspicious images using the quantization method, we can obtain statistically different from embedded-and-quantization attacked images and from quantization attacked-but-not-embedded sources. We have developed a technique based on one-class SVM for discriminating between cover-images and stego-images. Simulation results show our approach is able to distinguish between cover and stego images with reasonable accuracy.     

An atlas-based fuzzy connectedness method for automatic tissue classification in brain MRI

A framework incorporating a subject-registered atlas into the fuzzy connectedness (FC) method is proposed for the automatic tissue classification of 3D images of brain MRI. The pre-labeled atlas is first registered onto the subject to provide an initial approximate segmentation. The initial segmentation is used to estimate the intensity histograms of gray matter and white matter. Based on the estimated intensity histograms, multiple seed voxels are assigned to each tissue automatically. The normalized intensity histograms are utilized in the FC method as the intensity probability density function (PDF) directly. Relative fuzzy connectedness technique is adopted in the final classification of gray matter and white matter. Experimental results based on the 20 data sets from IBSR are included, as well as comparisons of the performance of our method with that of other published methods. This method is fully automatic and operator-independent. Therefore, it is expected to find wide applications, such as 3D visualization, radiation therapy planning, and medical database construction.     

Robust Multiclass Classification for Learning from Imbalanced Biomedical Data

Imbalanced data is a common and serious problem in many biomedical classification tasks. It causes a bias on the training of classifiers and results in lower accuracy of minority classes prediction. This problem has attracted a lot of research interests in the past decade. Unfortunately, most research efforts only concentrate on 2-class problems. In this paper, we study a new method of formulating a multiclass Support Vector Machine (SVM) problem for imbalanced biomedical data to improve the classification performance. The proposed method applies cost-sensitive approach and ramp loss function to the Crammer and Singer multiclass SVM formulation. Experimental results on multiple biomedical datasets show that the proposed solution can effectively cure the problem when the datasets are noisy and highly imbalanced.     

Rough set-based feature selection method

A new feature selection method is proposed based on the discern matrix in rough set in this paper. The main idea of this method is that the most effective feature, if used for classification, can distinguish the most number of samples belonging to different classes. Experiments are performed using this method to select relevant features for artificial datasets and real-world datasets. Results show that the selection method proposed can correctly select all the relevant features of artificial datasets and drastically reduce the number of features at the same time. In addition, when this method is used for the selection of classification features of real-world underwater targets,the number of classification features after selection drops to 20% of the original feature set, and the classification accuracy increases about 6% using dataset after feature selection.     

Fuzzy ARTMAP neural network for seafloor classification from multibeam sonar data

This paper presents a seafloor classification method of multibeam sonar data, based on the use of Adaptive Resonance Theory （ART） neural networks. A general ART-based neural network, Fuzzy ARTMAP, has been proposed for seafloor classification of multibeam sonar data. An evolutionary strategy was used to generate new training samples near the cluster boundaries of the neural network, therefore the weights can be revised and refined by supervised learning. The proposed method resolves the training problem for Fuzzy ARTMAP neural networks, which are applied to seafloor classification of multibeam sonar data when there are less than adequate ground-troth samples. The results were synthetically analyzed in comparison with the standard Fuzzy ARTMAP network and a conventional Bayesian classifier. The conclusion can be drawn that Fuzzy ARTMAP neural networks combining with GA algorithms can be alternative powerful tools for seafloor classification of multibeam sonar data.     

AdaBoost for Improved Voice-Band Signal Classification

A good voice-band signal classification can not only enable the safe application of speech ceding techniques, the implementation of a Digital Signal Interpolation （DSI） system, but also facilitate network administration and planning by providing accurate voice-band traffic analysis. A new method is proposed to detect and classify the presence of various voice-band signals on the General Switched Telephone Network （GSTN）. The method uses a combination of simple base classifiers through the AdaBoost algorithm. The conventional classification features for voice- band data classification are combined and optimized by the AdaBoost algorithm and spectral subtraction method. Experiments show the simpleness, effectiveness, efficiency and flexibility of the method.     

Study on optimization of agent initial positions in land combat simulation

The use of computational-intelligence-based techniques in the optimization of agent initial positions in land combat simulations is studied. A novel method for the reduction of support vectors in the support vector machine (SVM) is presented. The optimization on the width of the Gaussian kernel function and the combination of the SVM with the radial basis function neural network are performed in the proposed method. Simulation results show that the proposed method can improve the running efficiency drastically compared with that using the traditional SVM with the same precision. We also summarize and present some experiences and trends in the study on the optimization problem in land combat simulation.     

A multi-label classification algorithm based on label-specific features

Aiming at the problem of multi-label classification,a multi-label classification algorithm based on label-specific features is proposed in this paper.In this algorithm,we compute feature density on the positive and negative instances set of each class firstly and then select mk features of high density from the positive and negative instances set of each class,respectively;the intersection is taken as the label-specific features of the corresponding class.Finally,multi-label data are classified on the basis of label-specific features.The algorithm can show the label-specific features of each class.Experiments show that our proposed method,the MLSF algorithm,performs significantly better than the other state-of-the-art multi-label learning approaches.     

Support vector machine based on chaos particle swarm optimization for fault diagnosis of rotating machine

The performance of the support vector machine models depends on a proper setting of its parameters to a great extent. A novel method of searching the optimal parameters of support vector machine based on chaos particle swarm optimization is proposed. A multifault classification model based on SVM optimized by chaos particle swarm optimization is established and applied to the fault diagnosis of rotating machines. The results show that the proposed fault classification model outperforms the neural network trained by chaos particle swarm optimization and least squares support vector machine, and the precision and reliability of the fault classification results can meet the requirement of practical application. It indicates that chaos particle swarm optimization is a suitable method for searching the optimal parameters of support vector machine.     

Assessing Texture of Slub-Yarn Fabric Using Image Analysis

The application of digital image processing to the classification of the slub-yarn texture is discussed. Texture of the slub-yarn fabric is analyzed by using the texture analysis techniques. The influence of the slub-yarn parameters on the fabric texture is discussed. Results indicate that texture of the slub-yarn fabric can be reliably measured using gray level co-occurrence matrix （GLCM） analysis. The four indices of GLCM, the angular second moment, the contrast, the inverse difference moment and the correlation, are sensitive to the change of the slub-yarn parameters, and can be regarded as the major indices for the texture.     

New method of image retrieval based on interest image region blocks and comprehensive features

We propose an image retrieval method based on interest image region by asymmetrical blocking. An image is segmented into the interest region and background region on a certain rule. For the interest image regions, the color histogram of the uneven blocks is extracted as the color characteristic. We also collect the mean and variance value of the Gabor filtering results of background blocks as texture features of the background image. Then, the images can be retrieved by synthesizing the image color and texture features. We test our approaches by analyzing the results of recall and precision indicators for the Corel image data-base. The experiment results show that the proposed method performs effectively and accurately, which is more effective to retrieve the distant-view images, and the achieved precision increases by about 10% without loss of the retrieval call compared with some other traditional search methods.     

Online LS-SVM for function estimation and classification

An online algorithm for training LS-SVM (Least Square Support Vector Machines) was proposed for the application of function estimation and classification. Online LS-SVM means that LS-SVM can be trained in an incremental way, and can be pruned to get sparse approximation in a decremental way. When a SV (Support Vector) is added or removed, the online algorithm avoids computing large-scale matrix inverse. Thus the computation cost is reduced. Online algorithm is especially useful to realistic function estimation problem such as system identification. The experiments with benchmark function estimation problem and classification problem show the validity of this online algorithm.     

Three-phase Pupil Localization Method in Non-ideal Eye Images

Pupil localization is a very important preprocessing step in many reel applications. Accurate and robust pupil localization in non-ideal eye images is a challenging task. A detailed method of pupil localization in non-ideal eye images is proposed. This method is implemented in three main phases： first, segment the rough pupil region based on Gaussian Mixture Model： then modify the rough segmentation result using morphological method to minimize the influence of some disturbing factors; last estimate the pupil parameters based on minimizing the least square error. The proposed method is first tested on CASIA iris image dataset, and then on our self-captured iris dataset which contains a wider variety of iris images. Experiments show that the proposed method can perform well for nonideal eye images of various qualities.     

IBCIS: intelligent blood cell identification system

The analysis of blood cells in microscope images can provide useful information concerning the health of patients. There are three major blood cell types, namely, erythrocytes （red）, leukocytes （white）, and platelets. Manual classification is time consuming and suscep- tible to error due to the different morphological features of the cells. This paper presents an intelligent system that simulates a human visual inspection and classification of the three blood cell types, The proposed system comprises two phases： The image preprocessing phase where blood cell features are extracted via global pattern averaging, and the neural network arbitration phase where training is the first and then classification is carried out. Experimental results suggest that the proposed method performs well in identifying blood cell types regardless of their irregular shapes, sizes and orientation, thus providing a fast, simple and efficient rotational and scale invariant blood cell identification system which can be used in automating laboratory reporting.     

Anthropomorphic Classification of Tactile Qualities of Woven Fabrics Based on Skin/Textile Friction-Induced Vibrations

The common method classifying tactile qualities of fabrics is indirectly based on their difference of purely mechanical and physical properties. When human skin slides across fabric surfaces, the friction interaction between fabrics and skin will occur and trigger the cutaneouS tactile receptors, which are responsible for perceived tactile sensation. By the extracted features from friction- induced vibration signals, this paper presents an anthropomorphic classification method classifying tactile qualities of fabrics. The friction-induced vibration signals are recorded by a three-axis accelerator sensor, and the entice testing procedure is conducted in an anthropomorphic way to obtain vibration signals. The fast Fourier transform （FFT） is applied to analyzing the recoded signals, and then the classification features are extracted from the FFT data by the neurophysiological properties of tactile receptors. The extracted features are used to classify fabric samples by the softness sensation and the roughness sensation, respectively, and the classification performance is checked by a comparison with those in a sensory evaluation procedure. The results showed that the anthropomorphic objective classification method was precise and efficient to clarify tactile qualities of woven fabrics.