Search

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.

This paper aims to analyze computing performance of an on-board locomotion mode recognition system which was designed for robotic transtibial prosthesis. A hard-ware prototype based on field-programmable gate array was developed. Several algorithms, including support vector machine, back-propagation neural network, quadratic discriminant analysis and linear discriminant analysis, were implemented on board. Experiments on a transtibial amputee subject demonstrated that the proposed system can provide satisfactory acceleration effects on the four applied algorithms.

This paper presents an acceleration system for on-board support vector machine (SVM) model training for robotic transtibial prosthesis based on system on chip with field-programmable gate array (SoC-FPGA) hardware. A hardware prototype was developed and SVM-based model training algorithm was implemented with high-level synthesis technology. Experiments on a transtibial amputee subject demonstrated that the proposed system provided good speedups over ARM-based implementation for on-board training in six locomotion identification tasks (standing, level-ground walking, ramp ascent, ramp descent, stair ascent, stair descent). Meanwhile, the additional power consumption was not significant and acceptable.

Underwater visual understanding tends to be one of the most important challenges in ocean investigations recent years. In this paper, we make an attempt to develop a depth estimation approach from one single underwater image with non-uniform illumination correction. First, we try to remove the relatively strong reflection layer from the scene layer in those underwater images with non-uniform illumination and rely on the saturation detection to characterize the local regions of the artificial illumination in underwater images for compensation in depth estimation. Then we try to implement underwater dark channel prior and basically consider the blue and green color channels as the underwater visual source, and after one veilling light evaluation in the underwater scene layer, we further adopt one kind of fast guided filter into the rough transmission with low saturation compensation for illumination correction to refine a more dedicated depth estimation. In the simulation experiment, we propose that the proposed scheme can achieve good performance in ocean depth estimation with the help of uneven illumination correction.

The aim of this work was to improve the corrosion resistance of the cast ZM5 Mg alloy by laser surface alloying with Al powder. The laser alloying with Al on ZM5 Mg alloy was carried out by a pulsed Nd: YAG laser. The microhardness of the surface alloyed layer was measured by a microhardness Vickers tester. The microstructure of the alloyed layer is network. Microhardness of the laser surface alloyed layer was enhanced to 125–143Hv as compared to 72.7–92.6Hv for the untreated substrate. The corrosion resistance of the surface alloyed layer was improved as compared to alloy without laser treatment.

To study the beam instability caused by wake field effect, a dielectric wall accelerator module model consisting of symmetric pulse forming line, micro-stacked insulator and conductor layer was established. The finite integral method was used to simulate the wake field in the accelerating structure of dielectric wall accelerator (DWA) for high-current electron beam, wake potential and wake field impedance were calculated for the single and two accelerating units. Structural parameters of distance between two modules, thickness of metal and dielectric constant of insulator in micro-stacked insulator were changed to study the influence of different parameters on wake field. The results shows the wake potential and wake field impedance was small in x, y direction and different structural parameters can influence the wake field effect. Also it proved the advantage of accelerating cavity length and continuity in the DWA structure.

In orthogonal frequency division multiplexing (OFDM)-based cognitive radio (CR) systems, the sub-carriers, which span the spectrum used by licensed users (LUs), should be deactivated to avoid interference. The resulted OFDM system is called non-contiguous OFDM (NC-OFDM) system. For NC-OFDM systems, as no pilot symbols can be transmitted over these deactivated subcarriers, the conventional modified least square channel estimation can not be applied. In this paper, we propose a new channel estimation method to address this issue. In the proposed channel estimation, we use the LS channel estimation to obtain the channel fading coefficients associated with the subcarriers over which the pilot symbols are transmitted. Furthermore, we compute the time domain channel pulse response based on the basic principle of Fourier transform. Then, we adopt inverse discrete Fourier transform (IDFT) based channel estimation noise reduction to improve the channel estimation accuracy. For slow time-varying channel, we also propose to use pilot-block average based channel estimation noise reduction to further improve the channel estimation performance. Simulation results validate our proposed channel estimation method.

A pulsed gyrotron-based system has been set up in order to investigate millimeter wave sintering of ceramics. This paper describes the facility, including the gyrotron, propagation system, work chamber, and control system. An initial experiment on ceramic sintering has been completed for 98%Al/sub 2/O/sub 3/+1%SiO/sub 2/+1%MgO samples with the facility, which shows that 93.45% theoretical density can be reached for a sintering temperature of 1150/spl deg/C for 30 minutes.