Papers by Keyword: Feature Extraction

Abstract: Weld seam tracking is the key technology of industrial robot. It satisfies accuracy and real-time demand. This paper presents a robust recognition algorithm for weld seam images based on mathematical morphology. This algorithm used the morphological filtering for image enhancement at first, and then applied Otsu threshold segmentation in local. Besides, it implemented region filling and morphological thinning operation to get the segmented weld seam. Finally, the Hough transform was adopted to detect the straight lines from the weld image. The results prove that this method not only can detect the seam tracking in real-time, but also can identify weld seam automatically.

Abstract: In order to extract the fault feature validity in early fault diagnosis, method based on kernel principal component analysis and genetic programming (GP) is presented. The time domain features of the vibration signal are extracted and the initial symptom parameters (SP) are constructed. Then the combination to the initial SPs is carried on to optimize and build composite characteristics by GP. Through kernel principal component analysis (KPCA), the nonlinear principal component of the original characteristics is produced. Finally, the nonlinear principal components are selected as the feature subspace to classify the conditions of rolling bearing. Meanwhile, the within-class and among-class distance is introduced to compare and analyze the bearing condition recognition effect by using KPCA and GP plus KPCA separately. Experimental results show that the features extracted by kernel principal component analysis and genetic programming perform better ability in identifying the working states of the rolling bearing.

Abstract: This study reports about a comparison in recognizing road signs between Neural Network and Principal Component Analysis (PCA). The road sign with circular, triangular, octagonal and diamond shapes have been used in this study. Two recognition systems to determine the classes of the road signs class were implemented which are based on Feed Forward Neural Network and Principal Component Analysis (PCA). The performance of the trained classifier using Scaled Conjugate Gradient (SCG) back propagation function in Neural Network and PCA technique were evaluated on our test datasets. The experiments show that the system using PCA has a higher accuracy as compared to Neural Network with a minimum of 94% classification rate of road signs.

Abstract: This paper conducts a comprehensive research and discussion on the relevant technologies and manifold learning.Traditional MFCC phonetic feature will lead a slower learning speed on account of it has high dimension and is large in data quantities. In order to solve this problem, we introduce a manifold learning, putting forward two new extraction methods of MFCC-Manifold phonetic feature. We can reduce dimensions by making use of ISOMAP algorithm which bases on the classical MDS (Multidimensional scaling). Introducing geodesic distance to replace the original European distance data will make twenty-four dimensional data, which using the traditional MFCC feature extraction down to ten dimensional data.

Abstract: According to the high fault rate and the great difficulty of diagnosis for the automobile engine, an automobile engine faults detection system was designed. Because the vibration signal of the engine could reflect the faults types to a great extent, a fault detection method was proposed based on the extraction of the vibration signal correlation dimension. The collected vibration signal which was from different type of automobile engines was processed and analyzed. The correlation dimension was extracted and an improved correlation algorithm was proposed in the system, the computational accuracy was improved, and the standard deviation of the improved algorithm lowers about 50% in comparison with the traditional algorithm, the classification performance is raised variously, the excellent detection performance was showed in the system. The detection result shows that the correlation dimension feature extraction method that this paper proposed can detect and diagnose different types of automobile engine faults such as start subsystem fault, ignition subsystem fault, fuel supply subsystem, etc. The detection conclusion was stable and the simulation result has much great application performance.

Abstract: In video image sequences, assume that face forms larger interference in athletic process and use traditional algorithm to extract facial features which may lead to target pixel blending and feature missing problems. Three-dimensional face reconstruction has poor authenticity and characteristic distortion. In order to solve this problem, this paper proposes an anti-interference three-dimensional motion face feature extraction method based on multiple target constraint stereoscopic vision algorithm. Extract different facial images target feature points from video sequence, accurately calculate characteristics deformation in the process of face movement by adopting deformation constraint analysis method, resist the interferences of characteristics loss, and then make use of stereo vision technology to extract three-dimensional facial features. The experimental results show that this algorithm can effectively improve three-dimensional facial feature extraction in motion state, and achieve satisfactory results.

Abstract: The nonlinear feature of the rotating machinery signal in fault situation was extracted and researched based on the collected the vibration signal. And the extraction algorithm was researched. On the basis of the phase space reconstitution the recurrence plot (RP) algorithm was researched with the nonlinear time series analysis method. The inner feature of the recurrence plots was analyzed quantitatively, the feature called recurrence rate was extracted finally. Simulation result shows that the extracted feature has the function and property diagnosis of the machinery faults based on the RP and recurrence quantification analysis (RQA) methods and also with nice engineering application value.

Abstract: KPCA extracting principal component with nonlinear method is an improved PCA. The KPCA can extract the feature set which is more suitable in categorization than the conventional PCA. The method of KFDA is equivalent to KPCA plus LDA. KPCA is first performed and then LDA is used for a second feature extraction in the KPCA-transformed space. The KPCA and KFDA have been got widely used in feature extraction and face recognition. In this paper, the method of KPCA and KFDA is analyzed and their nature is revealed. Finally, the effectiveness of the algorithm is verified using the ORL database.

Abstract: For the purpose of extract the fault feature hidden by strong noise background in rolling bearing fault signal, a morphological undecimated wavelet method was proposed. A undecimated wavelet decomposition operator called gradient filter was presented based on an open-closed and closed-open mixed filter. The morphological gradient filter was used to extract the impulse feature of signal. The type and the length of structure elements used in these filters were alterable adapt to the signal. The method was applied to analyze the simulated data and measured vibration signals from the bearing with fault. The results confirm that the proposed method is feasible in impulse feature extraction of signal, and it is more effective than other traditional morphological undecimated wavelet methods.

Abstract: This paper presents a novel automatic algorithm for point cloud segmentation by using moving probability. An arbitrary point in point cloud is selected as the first seed point. Starting from the seed point, moving probability between the starting point and each of neighborhood points is estimated. Once one or more points with probabilities greater than a given threshold are identified, the starting point will move to these neighborhood points and new starting points are generated. Moving probabilities are estimated again and starting points move continually until all calculated probabilities are less than the threshold. Visited points are segmented from point cloud data. The second seed point is selected arbitrarily from the rest of points and the process is repeated. As a result, point cloud is segmented into individual feature regions. Experimental results show the effectiveness of the proposed algorithm.