Papers by Keyword: Feature Extraction

Abstract: In order to effectively extract nonstationary and nonlinear fault signature of hydropower units’ signals, a new method, based on Hilbert–Huang transform (HHT), is proposed. This method is used to carry out EMD (Empirical Mode Decomposition) analysis and Hilbert transform of signals firstly and then extract Hilbert spectrum to provide a basis for signal feature extraction. The vibration signal of upper brackets in hydropower units has been put forward with experimental analysis. The results suggest that the EMD can decompose vibration components in different frequency domain, which has intuitive physical meaning. Moreover, Hilbert spectrum also has a good resolution in time domain and frequency domain. Thus, HHT can be used to depict the fault signals effectively and lay the foundation of the fault pattern recognition.

Abstract: The active shape model (ASM) is a statistical parametric model, which is mainly used in image feature extraction. On the basis of the analysis of the original ASM texture model, a new texture modeling method was proposed in this paper. The improved method fully utilized the gray level information of adjacent points in the neighborhood of sampling points, improved the original face modeling method which only used the one-dimensional gray information with model matching accuracy problems. The experiments on Weizmann face database Indicate that, the improved method can obviously improve the feature point positioning accuracy, and accelerate the speed of face model fitting.

Abstract: An adaptive gender recognition method is proposed in this paper. At first, do multiwavlet transform to face image and get its low frequency information, then do feature extraction to the low frequency information using compressive sensing (CS), use extreme learning machine (ELM) to achieve gender recognition finally. In the process of feature extraction, we use genetic algorithm (GA) to get the number of measurements of CS in order to gain the highest recognition rate, so the method can adaptive access optimal performance. Experimental results show that compared with PDA and LDA, the new method improved the recognition accuracy substantially.

Abstract: In this paper, four original full typical UHF partial discharge signals are measured by using log-periodic antenna in 3-meter anechoic chamber environment. The relevance vector machine is applied for the study of partial discharge characteristics and the two relevance vector machine classifier are applied for the classification and identification of four partial discharge models. The experimental results are satisfactory. Compared with support vector machines, relevance vector machine can obtain more sparse classification model with probabilistic output value. It has a shorter test time and is more suitable for online testing. This method has a good prospect in partial discharge pattern recognition and online monitoring.

Abstract: Classification of moving military vehicle in battlefield is an important part of information acquirement. Support vector machine is a pattern classification method which is suitable to solve the small sample, non-linear classification problems. This paper uses one-versus-one multi-class SVM to classify military vehicle. This method is based on multi-sensor data including noise signal, the magnetic field disturbance signal, and vibration signal. The parameters of the SVM are determined by using the cross-validation method. The Simulation experiment results show that, compared to AdaBoost algorithm and two-class SVM, the one-versus-one multi-class SVM has higher accuracy.

Abstract: In modern digital era authentication has been done using biometric recognition. This biometric includes finger print, footprint, facial recognition, DNA of a person’s gene, hand palm print and eye’s iris recognition. The widely used among these is finger print and iris recognition. In this work we proposed a biometric recognition using footprints of a person. Earlier work deals with capturing footprint on a paper or on a surface. This won’t give us accurate foot print, since it depends on nature of the surface, quality of the paper and proper placement of the foot to give good foot print impression. To avoid all these we proposed a touch less method to obtain foot prints. The footprint can be obtained using any digital camera. We can take footprint image in many angles to conform the individuality of a person. In this work we used Principle Component Analysis (PCA) for pattern recognition and feature extraction. Then the SVM classifier split the patterns in to relevant classes. In early stage of our work itself we got remarkable quality and it is comparatively better than conventional footprint images obtained using paper or surface

Abstract: This paper presents wavelet based recognition of the machined surfaces namely turned, ground and shaped surfaces from the images acquired using Computer Vision System. Selection of mother wavelet has been done based on the peak signal to noise ratio (PSNR) value using Discrete wavelet transform (DWT) which has been used for feature extraction. Artificial neural network has been used to recognize the machined surfaces.

Abstract: Line spectrum is one kind of important feature information of the ship radiated noise, which provides a strong basis for the ship detection and identification. Due to the complexity of marine environment and the special nature of underwater acoustic channel,it is difficult to obtain ship radiated noise’ signals from the complex background noise. Under this situation, this paper proposed the cepstrum as the extraction method of obtaining feature information of the ship radiated noise, and the robust to noise performance was analyzed and compared. The analysis’ result shows that the method is simple, which is able to accurately extract feature line spectrum and shows good robust to noise.

Abstract: This paper focused on three common diseases, respectively, bacterial leaf spot, black spot and downy mildew as research targets, developed and designed a system to diagnose leaf diseases of sunflowers based on image identification. The system used MATLAB as platform and developed the system by utilizing GUI tool kit. Passing several tests, the system was believed to be able to identify three types of sunflower diseases effectively, respectively, bacterial leaf spot, black spot and downy mildew. The results basically met the requirements set before the design of this system.

Abstract: To match two or more images originated from the same scenario, a new fast automatic registration algorithm based on sparse feature point extraction is proposed. At the first step, the improved Harris corner detection algorithm is used to get two sets of feature points from the reference image and registration image. Second, a group of sparse feature points are selected from the reference image set as initial control points. Then, the corresponding matching points in the registration image set are searched based on local moment invariant similarity detection. Experimental results demonstrate that this method is fast and efficient.