Abstract In a myriad of engineering situations, we often hope to establish model which can acquire load conditions around structures through flow features detection. A data-driven method is developed predict the pressure on cylinder from velocity distributions in its wake flow. The proposed deep learning neural network constituted with convolutional layers and fully–connected layers: process information by extraction, are gathered fully-connected obtain coefficients. By comparing output data...

Image edge detection is the basis for precise positioning and accurate cutting of cloth pattern contours. Compared with commonly used traditional methods, Holistically-nested Edge Detection has clearer more continuous results including reduction false rate. However, this method extracts a coarser thick outline. In order to extract high-precision outline, clearly distinguish main body from background, provide convenience follow-up machine cutting, paper proposes an improved based on...

针对近邻传播(AP)聚类算法的计算复杂度和准确性,该文提出一种分层组合的半监督近邻传播聚类算法(SAP-SC)。算法引入分层聚类的思想,将一次AP聚类过程等分成若干层聚类,使得处理过程简单、易于实现;每层只关注聚类困难的数据点,并通过构造成对点约束和使用子簇标签映射进行半监督学习;基于组合提升的方法将各层聚类结果加权叠加,从而提升了算法的准确性能。理论分析和实验结果表明:算法在聚类准确性和计算复杂度方面有了较大改进。

Night vision technology is becoming ever-more widely used in military and civil fields, it will be more accurate for target detection recognition through color fusion of infrared low light level images. Since the classic Waxman model-only simulates rattlesnake's IR-depressed Visual Cell image not obvious, a novel model proposed this paper. We enhance edge information ON neural network light-level images then establish mathematical to process "enhanced cells"and "depressed cells". Next, we...

An image segmentation algorithm based on the ramp width reduction combined with an Intersecting Cortical Model (ICM) is proposed against problems that ICM in of weak edge produces geometric distortion. By virtue prewitt boundary operator and model, defines objective point, adjusts grey level pixel, reduces edge. On this basis, paper uses 2D histogram to expand cross entropy space so as obtain optical threshold ICM. The experiment indicates not only overcomes impact blur segment accurately,...

Autonomous Navigation Vehicle (AUV) navigates to the specified longitude and latitude on water surface by means of Global Positioning System (GPS), Inertial (INS) Electronic Compass (EC), then dives depth according measurements Depth Gauge (DG). To overcome low location accuracy existing navigation technologies in AUV, an integrated method based multi-sensor data fusion GPS/INS/EC/DG has been proposed. Since systematic stability high precision short-term measurement INS, it takes INS as...

Korean is an agglutinative language, in which pronunciations are affected by long-term context. In this paper, the long-time temporal information investigated to improve LVCSR. TRAP-based MLP features, able utilize scattered acoustic over several hundred milliseconds, employed obtain additional besides conventional cepstral features. contrast traditional phoneme set, consonants initial and final positions taken as same, a more specific set constructed via taking position dependent. broadcast...

The structures of mechanical parts are becoming more and complex with the continuous expansion design industry. In this article, profile points reduced first according to angle distance threshold. Based on minimal integral square error (ISE), Genetic algorithms (GAs) then used line segments circular as primitive form cell so that feature can be identified extracted for parts. It is shown genetic successful parts, although a large amount points, computation time longer, will considered in our...

An image segmentation algorithm based on the ramp width reduction combined with an Intersecting Cortical Model (ICM) is proposed against problems that ICM in of weak edge produces geometric distortion. By virtue prewitt boundary operator and model, defines objective point, adjusts grey level pixel, reduces edge. On this basis, paper uses 2D histogram to expand cross entropy space so as obtain optical threshold ICM. The experiment indicates not only overcomes impact blur segment accurately,...

To get better performance of fusion, accurate segmentation and suitable clarity measures always play a significant part in region-based image fusion. In this paper, novel method is proposed for visible infrared which can effectively address the issues inaccurate unsuitable measures. On one hand, by applying superpixel algorithm such as linear spectral clustering (LSC), more homogeneous regions be acquired fusion risk generating artifacts also decreased. other salient region detection...

A new method of edge detection based on adaptive oriented double opponent neurons is presented in this paper, considering that the homocentric receptive field lack directionality, and anisotropic ODOG model will be badly restrained during weighting process. To get relatively complete image edges, directional operators are introduced to choose Difference Gaussians (DOG) or orientations Oriented (ODOG) automatically. Compared with DOG methods, methods detect weak edges effectively better...

In this paper, we discuss two control schemes, based on (1) centrally controlkd region-growing and (2) iterative quadtree splitting, for incorporating knowledge-based processes into the- segmentation of texture images. An important feature these schemes is that knowledge about the nature images directly involved in partition process rather than being used afterwards to label resulting segments partition. Prototype systems which implemented automatic interpretation seismic sections are...