Scene classification is one of the most important issues in remote sensing image processing. To obtain a high discriminative feature representation for an to be classified, traditional methods usually consider densely accumulate hand-crafted low-level descriptors (e.g., scale-invariant transform) by encoding techniques. However, performance largely limited as they are not capable describing rich semantic information contained various images. alleviate this problem, we propose novel method...

To improve the discriminative and generalization ability of lightweight network for face recognition, we propose an efficient variable group convolutional called VarGFaceNet. Variable convolution is introduced by VarGNet to solve conflict between small computational cost unbalance intensity inside a block. We employ design our which can support large scale identification while reduce parameters. Specifically, use head setting reserve essential information at start particular embedding...

Face manipulation has shown remarkable advances with the flourish of Generative Adversarial Networks. However, due to difficulties controlling structures and textures, it is challenging model poses expressions simultaneously, especially for extreme at high-resolution. In this article, we propose a novel framework that simplifies face into two correlated stages: boundary prediction stage disentangled synthesis stage. The first models jointly via images. Specifically, conditional...

Abstract Long‐chain branched poly(butylene succinate) were synthesized through a two‐step process of esterification and polycondensation, using 1,2,4‐butanetriol (1,2,4‐BT) as long‐chain branching agent. The effect branches on the crystallization behaviors, rheological properties, tensile properties was investigated systematically. results differential scanning calorimetry polarized optical microscopy showed that with increasing 1,2,4‐BT segments, temperatures glass transition increase...

This letter proposes a novel rotation-invariant feature for object detection in optical remote sensing images. Different from previous features, the proposed matrix (RIM) can incorporate partial angular spatial information addition to radial information. Moreover, it be further calculated between different rings redundant representation of layout. Based on RIM, we propose an RIM_FV_RPP detection. For image region, first densely extract RIM features overlapping blocks; then, these are encoded...

Abstract Poly(butylene succinate) (PBS)/pristine raw multiwalled carbon nanotube (MWCNT) composites were prepared in this work via simple melt compounding. Morphological observations indicated that the MWCNTs well dispersed PBS matrix. Moreover, incorporation of did not affect crystal form as measured by wide‐angle X‐ray diffraction. The rheology, crystallization behaviors, and thermal stabilities PBS/MWCNT studied detail. Compared with neat PBS, into matrix led to higher complex viscosities...

Person re-identification (ReID) aims to predict whether two images from different cameras belong the same person. Due low image quality and variance in view point body pose, it remains a difficult task. To solve task, model is supposed appropriately capture features that describe regions for identification. With simple intuition explicitly incorporating ReID with part awareness could be beneficial learning more discriminative feature space, we propose segmentation as an assistant perception...

Knowledge distillation is an effective approach to leverage a well-trained network or ensemble of them, named as the teacher, guide training student network. The outputs from teacher are used soft labels for supervising new Recent studies \citep{muller2019does,yuan2020revisiting} revealed intriguing property that making serves good regularization From perspective statistical learning, aims reduce variance, however how bias and variance change not clear with labels. In this paper, we...

Image restoration aims to reconstruct degraded images, e.g., denoising or deblurring. Existing works focus on designing task-specific methods and there are inadequate attempts at universal methods. However, simply unifying multiple tasks into one architecture suffers from uncontrollable undesired predictions. To address those issues, we explore prompt learning in architectures for image tasks. In this paper, present Degradation-aware Visual Prompts, which encode various types of degradation,...

In this paper, we propose a novel network design mechanism for efficient embedded computing. Inspired by the limited computing patterns, to fix number of channels in group convolution, instead existing practice that fixing total numbers. Our solution based network, named Variable Group Convolutional Network (VarGNet), can be optimized easier on hardware side, due more unified schemes among layers. Extensive experiments various vision tasks, including classification, detection, pixel-wise...

Abstract Branched poly(butylene succinate) (PBS) copolymers were synthesized, from succinic acid (SA), 1,4‐butanediol (1,4‐BD), and 1,2‐octanediol (1,2‐OD) through a two‐step process containing esterification polycondensation, with different mole fractions of 1,2‐OD segments. The branched PBS characterized 1 H‐NMR, differential scanning calorimetry (DSC), wide angle X‐ray diffraction (WAXD), thermogravimetric analysis (TGA), dynamic rheological testing, tensile properties analysis. results...

It is a challenge to diagnose insulation condition of stator bars using an ultrasonic technique, due the complex structures and high attenuation insulation. In this paper, five low frequency transducers were used in pulse-echo configuration try test standby bar generator (300 MW, 18 kV). With available transducer, four types tested on bar. The short side was compared with large one slot. near exit from slot end bar, semiconductive coating covered uncoated one. having insulation-copper...

Cutting various patterns in elastomeric sheets has emerged as a simple yet effective way to fabricate soft mechanical metamaterials with unusual or auxetic properties. Upon stretching, the cut ligaments of metamaterial deform severely and much higher stress is induced around notch due concentration. This would limit stretchability even result fracture cut-mediated metamaterials. Optimization materials been rarely, if at all, explored, partially difficulty nonlinear optimization analysis...

When a soft mechanical metamaterial, consisting of regular array representative volume elements (RVEs), is stressed up to large strain, the delicately tailored behavior RVE does not prevail in metamaterial due boundary effect and manufacturing imperfections. A sheet comprising RVEs designed for snapping-back exhibits random snapping-through instability when uniaxially stretched. We conceptualize that loss representativeness can be avoided by introducing fiber reinforcement regulate...

Abstract In order to realize the storage of residual coal in goaf on flue gas power plant, adsorption characteristics nitrogen dioxide plant were studied. The Gaussian09 was used study process NO 2 molecules at density functional (DFT) B3LYP/6-311G level, and model by established. Different quantities obtained using orbital energy changes molecular bond length changes. According principle adsorption, is considered be physical with endothermic heat. On basis simulation, self-organized...

This paper presents a novel real-time routing protocol, called CBRR, with less energy consumption for wireless sensor networks (WSNs). End-to-End requirements are fulfilled speed or delay constraint at each hop through integrating the contention and neighbor table mechanisms. More precisely, CBRR maintains via mechanism being dependent on broadcast instead of beacons. Comprehensive simulations show that can not only achieve higher performance in static networks, but also work well dynamic networks.

Face manipulation has shown remarkable advances with the flourish of Generative Adversarial Networks. However, due to difficulties controlling structures and textures, it is challenging model poses expressions simultaneously, especially for extreme at high-resolution. In this paper, we propose a novel framework that simplifies face into two correlated stages: boundary prediction stage disentangled synthesis stage. The first models jointly via images. Specifically, conditional encoder-decoder...

Abstract Scene classification for remote sensing is a popular topic, and many recent convolutional neural networks (CNNs)‐based methods have shown the great model capacity learning ability of highly discriminative features. Given large number training data, CNN can extract extensive features learn to predict image. However, supervised tasks, deep models often rely on labelled images, which are difficult pre‐process. Thus, lightweight essential. Easy‐classified hard samples may also cause an...

In order to further utilize the transmission line cooridor, meanwhile, achieve a relatively friendly electromagnetic environment, composite cross arm is developed and employed in system. The electric field distribution as well electromagentic environment of deserves detailed in-depth study verify its performance. this research, three-dimension (3D) finite element method (FEM) was calculate arm. characterized by intensity on ground, radio interference audible noise these parameters were...