Multiple Scale Features Research Articles

Pedestrian Retrieval Using Generated Samples and Multistream Layer in Sensor Networks

In this paper, we propose a novel loss function named the hybrid quadruplet loss (HQL) to utilize the generated samples for pedestrian retrieval in sensor networks. The proposed HQL employs a set of quadruplets in order to maintain an appropriate margin between the real sample and the generated sample, reduce the intra-class variations and enlarge the inter-class variations. By this way, the generalization of the deep model could be improved. Furthermore, to identify the extremely similar pedestrians, we propose a novel multistream layer to mine imperceptible information from different aspects. The proposed multistream layer utilizes various filters with different morphologies to capture discriminative features in multiple scales, and it is flexible to follow any convolutional layer. Experiments on the three large-scale pedestrian retrieval databases (Market1501, CUHK03, and DukeMTMC-reID) demonstrate that the proposed method outperforms other state-of-the-art methods.

Passion fruit detection and counting based on multiple scale faster R-CNN using RGB-D images

The accurate and reliable fruit detection in orchards is one of the most crucial tasks for supporting higher level agriculture tasks such as yield mapping and robotic harvesting. However, detecting and counting small fruit is a very challenging task under variable lighting conditions, low-resolutions and heavy occlusion by neighboring fruits or foliage. To robustly detect small fruits, an improved method is proposed based on multiple scale faster region-based convolutional neural networks (MS-FRCNN) approach using the color and depth images acquired with an RGB-D camera. The architecture of MS-FRCNN is improved to detect lower-level features by incorporating feature maps from shallower convolution feature maps for regions of interest (ROI) pooling. The detection framework consists of three phases. Firstly, multiple scale feature extractors are used to extract low and high features from RGB and depth images respectively. Then, RGB-detector and depth-detector are trained separately using MS-FRCNN. Finally, late fusion methods are explored for combining the RGB and depth detector. The detection framework was demonstrated and evaluated on two datasets that include passion fruit images under variable illumination conditions and occlusion. Compared with the faster R-CNN detector of RGB-D images, the recall, the precision and F1-score of MS-FRCNN method increased from 0.922 to 0.962, 0.850 to 0.931 and 0.885 to 0.946, respectively. Furthermore, the MS-FRCNN method effectively improves small passion fruit detection by achieving 0.909 of the F1 score. It is concluded that the detector based on MS-FRCNN can be applied practically in the actual orchard environment.

Multi-Scale Point-Wise Convolutional Neural Networks for 3D Object Segmentation From LiDAR Point Clouds in Large-Scale Environments

Although significant improvement has been achieved in fully autonomous driving and semantic high-definition map (HD) domains, most of the existing 3D point cloud segmentation methods cannot provide high representativeness and remarkable robustness. The principally increasing challenges remain in completely and efficiently extracting high-level 3D point cloud features, specifically in large-scale road environments. This paper provides an end-to-end feature extraction framework for 3D point cloud segmentation by using dynamic point-wise convolutional operations in multiple scales. Compared to existing point cloud segmentation methods that are commonly based on traditional convolutional neural networks (CNNs), our proposed method is less sensitive to data distribution and computational powers. This framework mainly includes four modules. Module I is first designed to construct a revised 3D point-wise convolutional operation. Then, a U-shaped downsampling-upsampling architecture is proposed to leverage both global and local features in multiple scales in Module II. Next, in Module III, high-level local edge features in 3D point neighborhoods are further extracted by using an adaptive graph convolutional neural network based on the K-Nearest Neighbor (KNN) algorithm. Finally, in Module IV, a conditional random field (CRF) algorithm is developed for postprocessing and segmentation result refinement. The proposed method was evaluated on three large-scale LiDAR point cloud datasets in both urban and indoor environments. The experimental results acquired by using different point cloud scenarios indicate our method can achieve state-of-the-art semantic segmentation performance in feature representativeness, segmentation accuracy, and technical robustness.

Accurate EMC Engineering on Realistic Platforms Using an Integral Equation Domain Decomposition Approach

This article investigates the efficiency, accuracy and versatility of a surface integral equation (SIE) multisolver scheme to address very complex and large-scale radiation problems including multiple scale features, in the context of realistic electromagnetic compatibility (EMC)/electromagnetic interference (EMI) studies. The tear-and-interconnect domain decomposition (DD) method is applied to properly decompose the problem into multiple subdomains attending to their material, geometrical, and scale properties, while different materials and arbitrarily shaped connections between them can be combined by using the so-called multiregion vector basis functions. The SIE-DD approach has been widely reported in the literature, mainly applied to scattering problems or small radiation problems. Complementarily, in this article, the focus is placed on realistic radiation problems, involving tens of antennas and sensors and including multiscale ingredients and multiple materials. Such kind of problems are very demanding in terms of both convergence and computational resources. Throughout two realistic case studies, the proposed SIE-DD approach is shown to be a powerful electromagnetic modeling tool to provide the accurate and fast solution which is indispensable to rigorously accomplish real-life EMC/EMI studies.

Outdoor Scene Understanding Based on Multi-Scale PBA Image Features and Point Cloud Features.

Outdoor scene understanding based on the results of point cloud classification plays an important role in mobile robots and autonomous vehicles equipped with a light detection and ranging (LiDAR) system. In this paper, a novel model named Panoramic Bearing Angle (PBA) images is proposed which is generated from 3D point clouds. In a PBA model, laser point clouds are projected onto the spherical surface to establish the correspondence relationship between the laser ranging point and the image pixels, and then we use the relative location relationship of the laser point in the 3D space to calculate the gray value of the corresponding pixel. To extract robust features from 3D laser point clouds, both image pyramid model and point cloud pyramid model are utilized to extract multiple-scale features from PBA images and original point clouds, respectively. A Random Forest classifier is used to accomplish feature screening on extracted high-dimensional features to obtain the initial classification results. Moreover, reclassification is carried out to correct the misclassification points by remapping the classification results into the PBA images and using superpixel segmentation, which makes full use of the contextual information between laser points. Within each superpixel block, the reclassification is carried out again based on the results of the initial classification results, so as to correct some misclassification points and improve the classification accuracy. Two datasets published by ETH Zurich and MINES ParisTech are used to test the classification performance, and the results show the precision and recall rate of the proposed algorithms.

Fully Convolutional Single-Crop Siamese Networks for Real-Time Visual Object Tracking

The visual object tracking problem seeks to track an arbitrary object in a video, and many deep convolutional neural network-based algorithms have achieved significant performance improvements in recent years. However, most of them do not guarantee real-time operation due to the large computation overhead for deep feature extraction. This paper presents a single-crop visual object tracking algorithm based on a fully convolutional Siamese network (SiamFC). The proposed algorithm significantly reduces the computation burden by extracting multiple scale feature maps from a single image crop. Experimental results show that the proposed algorithm demonstrates superior speed performance in comparison with that of SiamFC.

Matching Algorithm of 3D Point Clouds Based on Multiscale Features and Covariance Matrix Descriptors

The three-dimensional (3D) point cloud is one of the most promising tools for representing and identifying 3D objects. The critical step for matching is to find the appropriate feature descriptors. Two prevalent descriptors are global feature descriptor and local feature descriptor. The former represents the geometric and topological properties of the neighborhood in the entire 3D model, but it can not recognize the covered areas. The local descriptor focuses on narrow neighborhoods, while coarse areas are still present for disambiguation. In this paper, we present a novel matching algorithm of 3D point clouds based on multiple scale features and covariance matrix descriptors. By the combination of the curvature and eigenvalue variation, the key points are detected precisely under multiple scales. Furthermore, we develop a three-scale covariance matrix descriptor to demonstrate local features of the key points. The three-scale covariance matrix descriptor includes the geometric angles, dimensionality, the ratio of projection length and the difference of the curvature, which can describe the local geometric features of the key points more clearly and make feature descriptors more distinguished, especially for key points which are similar in a small range but are not similar in a large range. Besides, a bidirectional proportion strategy is used to find the optimal matching pairs. The algorithm efficiently reduces the mismatching error compared with some local descriptor. Moreover, it is more robust to high noise. Experiments show the efficiency and the robustness of the proposed algorithm for matching three-dimensional point clouds with Gaussian noise and deformed shapes.

FSNet: A Target Detection Algorithm Based on a Fusion Shared Network

Currently, in a target detection framework, a feature pyramid is widely used to capture differences resulting from the change in size of a detected object. Examples include the one-stage detector, the Deconvolutional Single-Shot Detector (DSSD), the RefineDet and the two-stage detector DetNet. Although these target detection frameworks with characteristic pyramid structure have achieved good results, they are limited by the facts that 1) they only use one feature layer of multi-scale information for prediction, and 2) they do not fully incorporate multi-scale inclusion of feature information. Therefore, based on Single-shot detector (SSD) architecture, we propose a novel one-stage target detector framework-fusion shared network (FSNet) which makes full use of feature information between multiple scale feature maps to more effectively detect objects of different scales. First, we introduce a fusion attention mechanism to fuse features between feature layers of different scales. Secondly, we combine this with multi-head structure fusion multi-layer features extracted as new features by a backbone network. Finally, we feed multi-layer fusion features into shared prediction module (shared PM). In the module, a new class feature pyramid for target detection is formed wherein the feature map used for prediction is composed of multiple levels of feature layers. In order to verify the effectiveness of this new FSNet target detection algorithm, we used the 320X320 input for FSNet, obtaining 80.7% mean accuracy (mAP) and 15 frames per second (Frame Per Second, FPS) in the PASCAL VOC 2007 test.

Multiple Features Decomposition for Subcutaneous Vein Extraction and Measurement

Subcutaneous veins should be accurately segmented to satisfy the emerging demands in biometrics-based identity recognition, automated intravenous injection, and vascular surgical navigation. This paper proposes a novel segmentation method for extracting vascular vein structures from near-infrared images based on multiple feature classification. First, the isotropic undecimated wavelet transform in combination with the Hessian function is used to enhance the visibility of vascular structures. An approximated vascular structure is extracted by multiple-scale features based on the enhanced image. The establishment of the mapping relationship between the sketch image and source image spaces is assisted by mending and fitting procedures. Vascular features are also extracted from the enhanced contour image by a matched filter. Furthermore, the image can be classified into vascular, skin, and blurred regions by a K-means algorithm. The final vascular segmentation outcome is a refined structure. Segmentation results demonstrated that the proposed method is robust for subcutaneous vein images with different blur scales, where the completeness and smoothness of the extracted vein structure are retained. The proposed method is accurate and completely automatic, with potential applications in biometric feature recognition and intravenous injection.

Detecting fine scratches on smooth surfaces with multiscale wavelet representation

This paper presents a set of image-processing algorithms for automatic detection of fine scratches on smooth surfaces, such as automobile paint surfaces. The scratches to be detected have random directions, inconspicuous gray levels and background noise. The multiscale wavelet transform was used to extract texture features, and a controlled edge fusion model was employed to merge the detailed (horizontal, vertical and diagonal) wavelet coefficient maps. Based on the fused detail map, multivariate statistics were applied to synthesize features in multiple scales and directions, and an optimal threshold was set to separate scratches from the background. The experimental results of 24 automobile paint surface showed that the presented algorithms can effectively suppress background noise and detect scratches accurately.

Wavelet-based rotational invariant roughness features for texture classification and segmentation

In this paper, we introduce a rotational invariant feature set for texture segmentation and classification, based on an extension of fractal dimension (FD) features. The FD extracts roughness information from images considering all available scales at once. In this work, a single scale is considered at a time so that textures with scale-dependent properties are satisfactorily characterized. Single-scale features are combined with multiple-scale features for a more complete textural representation. Wavelets are employed for the computation of single- and multiple-scale roughness features because of their ability to extract information at different resolutions. Features are extracted in multiple directions using directional wavelets, and the feature vector is finally transformed to a rotational invariant feature vector that retains the texture directional information. An iterative K-means scheme is used for segmentation, and a simplified form of a Bayesian classifier is used for classification. The use of the roughness feature set results in high-quality segmentation performance. Furthermore, it is shown that the roughness feature set exhibits a higher classification rate than other feature vectors presented in this work. The feature set retains the important properties of FD-based features, namely insensitivity to absolute illumination and contrast.

Porous media: A model system for the physics of disorder

We discuss several examples of applications of soft matter and statistical physics to various problems related to porous and fractured media. The structure of porous media often displays multiple scale features which can be analysed by neutron diffraction or electron microscopy and can be approached by fractal models. Such characteristics are also observed on the surface of natural fractures.The relation between various transport parameters such as permeability or conductivity introduce characteristic microscopic length scales which can sometimes be independently determined. In some cases, if the porous medium get clogged or if the number of flow channels or fractures is low enough so that threshold effects appear which can be analysed in terms of percolation models. Disorder physics approaches are particularly useful to analyse non miscible diphasic flows in some cases in which multiscale heterogeneities of the fluid mixture composition appear. This is for instance the case for very slow non wetting invasions and of the fast injection of a low viscosity fluid : these processes can be described respectively by the “invasion percolation” and the “diffusion limited aggregation” models. Finally tracer dispersion provides an application of random walk models to disordered systems : examples of the response of this measurement in partly saturated and double porosity media are presented.

Upwind method for simulation of viscous flow on adaptively refined meshes

A new node-based upwind scheme for the solution of the three-dimensional Navier-Stokes equations on adaptively refined meshes is presented. The method uses a second-order upwind total variation diminishing scheme to integrate the convective terms and discretizes the viscous terms with a new compact central difference technique. Grid adaptation is achieved through directional division of hexahedral cells in response to evolving features as the solution converges. The method is advanced in time with a multistage Runge-Kutta time stepping scheme. Two- and three-dimensional examples establish the accuracy of the inviscid and viscous discretization. These investigations highlight the ability of the method to produce crisp shocks, while accurately and economically resolving viscous layers.. The representation of these and other structures is shown to be comparable to that obtained by structured methods. Further three-dimensional examples demonstrate the ability of the adaptive algorithm to effectively locate and resolve multiple scale features in complex three-dimensional flows with many interacting, viscous, and inviscid structures.