Search

In this work, we propose a new method termed Semantic Frustum Based VoxelNet (SFB-VoxelNet) for 3D object detection. Since previous frustum-based methods are greatly affected by the accuracy of 2D image proposal, we directly alleviate this influence by reducing the probability of omissions and reinforcing the discriminating ability of the 3D detector. Specifically, first, we use semantic map, generating by the method of semantic segmentation, to keep up semantic information and stuff of interest. Second, by projecting the semantic map to LIDAR space, the frustum is generated and then sent to our detector to estimate 3D object classification and bounding box location. Our method not only cuts down most of search space but also assigns a reasonable arrangement for the two steps of frustum-based methods. In addition, our method is evaluated on KITTI 3D object detection benchmark to verify its superiority while maintaining a real-time inference speed, comparing to other state-of-the-art literatures.

Normally-off Gallium Nitride (GaN) transistors with p-GaN gated technology for power applications are studied for the optimization of off-state breakdown performance. The gate-drain distance, gate length, field plate length and its configuration have been studied. Increase gate-drain distance will not only enhance the breakdown performance but also increase the $\mathrm{R}_{\mathrm{on}},\ \mathrm{L}_{\mathrm{GD}}$ between $10 -20\ \mu\mathrm{m}$ is recommended. Extra gate length will lead to the rising of the gate leakage current under high voltage. The overhang length of the field plate is a critical factor for breakdown performance, over placing the length of overhang may result in an additional current leakage path.

In this paper, we investigate a novel learning-based method for dynamic scene deblurring. Since the inference model is formulated as an encoder-decoder, the core task has turned to learning blur-invariant hidden features from the blurred images to a great degree. To achieve state-of-the-art results in terms of both deblurring accuracy and efficiency, a simple, robust and computationally efficient deep auto-encoder is developed tailored specifically for blind deblurring, which is learned in an adversarial fashion based on use of the recent Wasserstein generative adversarial networks. Thanks to the designed framework, the new model has shown comparable or superior performance both qualitatively and quantitatively to existing state-of-the-art methods. What is more important, instead of exploiting the multi-scale strategy as previous methods, our model is just single-scale capable of achieving 6 times more efficiency than the closest competitor by Nah et al. [1], which is a more complex multi-scale deep method.

In this paper, we propose a video enhancement method using temporal-spatial total variation Retinex and luminance adaption. To utilize the temporal information between video frames, we construct a illumination data fidelity term and propose an temporal-spatial total variation model for Retinex. In order to further enhance the contrast of video frames, we use the adaptive Gamma correction with weighting distribution as a post processing step. Thus, the proposed method is able to enhance video frames' contrast while producing coherent illumination. Experimental results demonstrate the efficiency of the proposed method.

In order to attain “peak load shifting”, the main problem is to solve operational optimal control of heat storage boiler in this paper. Making the electric heat storage boiler the main object of the thesis, considering constraints of electric heat storage boilers such as heat storage capacity, power and temperature, etc., the control strategy of electric heat storage boiler based on multi-agent is proposed through incorporating the ideas of multi-agent. With simulating the heating system of the electric heat storage boilers and regarding a single electric heat-storage boiler as a single-agent, the load demand model of electric heat storage boiler is built and multiple electric boilers can communicate and coordinate mutually, thus the optimization control of electric heat storage boiler is realized. Simulation for a specific case is conducted, and the results indicate that the proposed control strategy based on multi-agent can effectively achieve “peak load shifting” by the electric heating units, reducing the difference between peak and valley.

This paper introduces a fast blind deconvolution strategy for image deblurring by modifying a recent natural image model, i.e., the total generalized variation (TGV), which aims at reconstructing an image with higher-order smoothness as well as sharp edges. But, when it turns to the blind issue, as demonstrated either empirically or theoretically by a few previous blind deblurring works, natural image models including TGV are more often than not inclined to trivial solutions, e.g., the delta blur kernel and the input blurred images. Inspired by the discovery, a simple, yet effective modifying strategy is applied to the second-order TGV, leading to a novel l 0 –l 1 -norm-based image regularization adaptable to the blind deblurring problem. Then, a numerical algorithm is deduced with O(NlogN) complexity, via borrowing ideas of the operator splitting method, the augmented Lagrangian, and also the fast Fourier transform (FFT). Experimental results on both synthetic and real color blurred images demonstrate the superiority or comparable performance of the new approach to state-of-the-art ones, in terms of deblurring accuracy and efficiency.

Learning-based methods have been the mainstream in single image super-resolution (SISR) over the past decade or so. This paper, however, returns to the reconstruction perspective to SISR, and studies the possibility of achieving competitive or even better SISR performance compared with state-of-the-art learning-based approaches in the literature. Specifically, through the (fast) iterative shrinkage-thresholding algorithm, a general, simple, yet accurate SISR framework is proposed by embedding off-the-shelf image denoising algorithms into the reconstruction process rather than struggling to develop advanced image priors. Numerous experiments on the benchmark dataset are performed accompanied by comparisons against a few recent learning-based methods, e.g., A+, SRCNN, etc. The results have demonstrated well the rationality and superiority of the proposed framework in terms of the super-resolution accuracy as image filtering schemes such as BM3D, WNNM, and other possible candidates, are used. Note that, learning-based filtering schemes can be also plugged into our framework, which hence leads to hybrid reconstruction/learning-based SISR approaches, demonstrating the flexibility as well as the potential of the proposed framework.

In this paper, we studied the current carrying capacity of solder joints in the Copper Column Grid Array (CuCGA) by means of finite element method (FEM). The current carrying capacity of solder joints can be reflected by the peak value of the temperature and current density. And the solder fillets were the critical region in the entire CuCGA. Therefore, the paper mainly studied the influence of the shape-parameters for solder joints and wirings to the peak value of the temperature and current density. The results were as follows, the peak value of the current density of solder fillets occurred in the geometric discontinuity region connected the solder fillets and wirings. And the peak value of the temperature of soldering fillets occurred in the geometric discontinuity region connected the solder fillets and the copper columns. With the increase of height for solder columns, the peak value of current density was constant, while the peak value of temperature decreased slightly. With the increase of diameter for solder columns, both the peak value of current density and the temperature decreased. With the increase of climbing height for soldering, the current density in the solder joints decreased. With the increased of minimum width of wiring, both the peak value of temperature and the peak value of current density appeared the trend of decreased.

Nanoindentation experiment together with the finite element simulation was used to study the mechanical behavior of Sn3.0Ag0.5Cu (SAC305) lead-free solder joint under cyclic loading-unloading. The effect of different maximum loads and holding time on the micromechanics behavior was investigated. The results indicated that the increase of the maximum load caused the increase of the depth of residual indentation and the area of hysteresis loop. The creep displacement of solder joint was increased with the prolonging of holding time. It increased sharply at first, and then went stabilized. In addition, a good agreement between finite element model and experimental results was reached.

The nonparametric blind deconvolution, either single or multi-shot, has been intensively studied since Fergus et al.'s variational Bayesian approach to camera shake deblurring [1]. However, in the current literature there is always a gap between the two highly related problems; single and multi-shot blind deconvolution are modeled and solved independently, lacking a unified optimization perspective. We attempt to bridge the gap between the two problems in this paper, via constructing a simple, rigorous and unified minimization functional which couples the variational Bayesian and maximum-a-posteriori principles. The new approach is depicted using a directed graphic model, in which the sharp image and the inverse noise variance associated with each shot are treated as random variables, while the blur-kernel, in difference from existing variational Bayesian methods, is just modeled as a deterministic parameter. With a universal, three-level hierarchical prior on the latent sharp image and a Gamma hyper-prior on each inverse noise variance, the single/multi-shot blind deconvolution is formulated into an l0.5-norm regularized negative log-marginal-likelihood minimization problem. By ideas of expectation-maximization, majorization-minimization, mean field approximation, and iteratively reweighted least squares, all the unknown quantities of interest, including the sharp image, the blur-kernel, the inverse noise variance, as well as other related parameters are estimated automatically. In comparison to existing single/multi-shot methods, the proposed method is not only more flexible, but also more adaptive while with less implementational heuristics. Experimental results on Levin et al.'s [2] benchmark data set demonstrate the effectiveness and superiority of the proposed framework.

Single image blind deblurring has been intensively studied since Fergus et al.'s variational Bayes method in 2006. It is now commonly believed that the blur-kernel estimation accuracy is highly dependent on the pursed salient edge information from the blurred image, which stimulates numerous l0-approximating blind deblurring methods via kinds of techniques and tricks. This paper, however, focuses on the four recent daring attempts which are all based on the simple and direct l0-norm. A systematic comparative analysis is made towards those methods, clarifying their similarities and differences, and providing a benchmark evaluation on both the deblurring quality and computational efficiency. Results have demonstrated that the l0-norm alone is far enough to achieve top blind deblurring performance. Instead, details are to be paid with fairly more attention as working on the problem formulation as well as the algorithmic deduction. Besides, three possible extensions are analyzed on the most leading bi-l0-l2-norm regularization-based blind deblurring method.

We propose a structured feature active contour model based on the level set method for image segmentation. We make the following two contributions. First, an adaptive data fitting term detects intensity variation based on the direction and global information. Second, integrated with a structured gradient vector flow (SGVF) method, we formulate a new regularization term with respect to the level set function via the duality formulation to penalize the length of active contour. We compare the proposed method to the classical active contour methods and demonstrate through the experiments on synthetic and medical images.

The GIS-based highway rock slope stability evaluation expert system platform is established. The characteristics of road projects are long tracks; plenty work sites; more complex conditions; dynamic construction information and other technical difficulties. The system platform is effective way to assess the rock slope stability. The GIS system as a basic platform, the implementation of rock cutting slope of three-dimensional virtual space is carried out, completing obtaining the on-site slope topography, geology and other geological information at the same time. Then the various types of slope data is stored and updated conveniently, a fast and practical the rock slope stability assessment system for rock slope is put forward, which could provide decision-making basis to the disaster prediction and prevention.

This paper presents a new digital image stabilization framework using an improved point matching algorithm. The point in the image is chosen using a lattice pattern, because the lattice can simulate the different transform model. And the motion is estimated by matching the lattice. This algorithm is compatible with different motion models according to the real experimental results. In the situation of translation and rotation, the coordinates of the transform can be calculated and stored in advance to reduce the amount of computation.

Images acquired by an optical system are usually degraded by atmospheric turbulence, which exists in the path between the targets and the imaging system. The estimation of atmospheric turbulence degraded Point Spread Function (PSF) without any prior knowledge of clear images is the most challenging and significant technique on image restoration. In this paper, a new PSF estimation approach is proposed for long-exposure atmospheric turbulence degraded images, and is applied to image restoration successfully. The PSF is estimated via an isosceles model which is proposed to approximate one component of the original image's Fourier amplitude. On degraded image restoration, the short-exposure image frames are transformed into a single long-exposure image, and then the restored image is obtained by using the estimated PSF and Wiener Filter. Numerical experiments suggest that this algorithm can obtain accurate PSF from both synthetic and real images. It is also shown objectively that the quality of restored images is greatly enhanced, by applying the Gray Mean Grads and Laplacian Sum standards.

In this paper, a new region-based active contour model is proposed for image segmentation. Different from the general region-based active contour models, this model partitions the regions of interests in images depending on the local statistics of the intensity and the magnitude of gradient in the neighborhood of the contour. Inspired by the structure tensor method, an improved regularization term is defined through the duality formulation to penalize the length of region boundaries. Experiments on medical images demonstrate the proposed model outperforms the classical segmentation models in terms of efficiency and accuracy.

The large crude oil storage tank with a capacity of 10×104 m3 has been widely used in China currently. In order to obtain the stress distribution in the wall board, bottom board, big fillet weld and manhole, and to verify the rationality of tank design, a stress test which combining the method of static resistance strain test (SRST) and optical fiber strain test (OFST) is conducted during water filling test for the tank. Software ANSYS is employed to construct a three-dimensional (3d) space finite element model under hydraulic test loading. And then theoretically calculated and measured values are compared and a good agreement between them is achieved.

In order to research the shallow landslide hazards of soil slopes in seasonal frozen regions, the critical depth analysis methods are put forward. Continuum explicit finite difference method is applied in the research. Considering the temperature, gravity and seepage force, the soil slope solid model of seasonal frozen regions is established. As the outside temperature alter from −5°C to 10°C, then the unfavorable stress position and failure depth of the fixed water content soil slope of seasonal frozen regions is put forward. Based on the critical depth experiment data of freeze-thawing surface, the safety factor variations of different water content and F-T times are compared. Analysis reveals that, the failure surface is the interface between the frozen and thawed surface in the slope. With the increasing of water content and F- T times, the critical depth migrates to the surface gradually, and finally reaches to the average maximum depth of the region. The conclusion could present more reference of the slope stability analysis and supervision in the seasonal frozen area.

In order to study the effect of backlash nonlinear on resonance frequency of electric actuator transmission system, in the premise of meeting the stability, it establishes balanced equation of dynamics system with backlash, and does some theory research and dynamic analysis of backlash nonlinearity and system natural frequency and resonance frequency. First, according to describing function to theoretically model the backlash nonlinear, the relation between describing function and the amount of backlash and input amplitude is analyzed; then, the dynamic equilibrium equation of transmission system with backlash is established, the effect of backlash on system natural frequency and resonance frequency; Finally, it makes dynamic simulation analysis for the transmission system based on virtual prototype technology, and the theoretical values and simulation values are compared. Theoretical analysis and simulation results show that the theoretical analysis is basically consistent with simulation results when it is zero backlash, and the error of theoretical analysis and simulation results is within 10%. Therefore, it can be concluded that considering the transmission system theoretical analysis of backlash nonlinearity is feasible, and it is much closer to the actual system with instructive significance.

The imperfections in the radio-frequency coils or problems associated with the acquisition sequences may cause MRI intensity inhomogeneities, which may mislead image segmentation. Comparing the tradition fuzzy C means model, this paper adds the bias field information in the objective function for simultaneous correction of the bias field and accurately segmentation. In adaptive model, the bias field is modeled as a linear combination of a set of basis functions to ensure the smoothness and slowly varying, and can be easy estimated by computing the coefficients of this basis functions in every iteration. Experimental results on both synthetic images and MR data are given to demonstrate the effectiveness and efficiency of the proposed algorithm.

Scattering centers estimation results are significant for radar target recognition in high frequency section. This paper introduces biquaternion theory into full-polarization scattering centers extraction of radar target. Firstly, we define full-polarization scattering centers as a biquaternion model. It makes a close connection between polarization parameters. Then, we propose Bihan’s BQ-MUSIC thought to extract scattering centers’ position parameters. Simulations illustrate that our theory gets better estimation results compared to polarization MUSIC algorithm of complex model.

In this paper, designed the slip motor speed control system for packing rubber strip which is used in plastic steel door and window. The system, 32-bit embedded microprocessor LPC2136 as the control core, PWM speed control, velocity feedback from tachogenerator, rubber strip position feedback from the CMOS camera, has proved that CMOS camera's image quality meets the engineering requirement; Slip motor speed is in the precision of 1 percent in scope of 0 to 1400rpm. Practical application shows that the system is reliable, stable and higher precision of control. The system design provides reference for similar control system and builds the basis for the design of intelligent control system.

In order to enhance the resolution of Charge Coupled Device(CCD)pixel with the pixel dimension and focal length of optical system unchanged, the problem of CCD geometric superresolution is educed. The problem is discussed in terms of software and hardware. In software, the bicubic interpolation is introduced, while in hardware, the subpixel imaging method is utilized. The subpixel imaging method integrates two uniform linear CCDs staggered by half dimension of pixel in one device. In operation, the line frequency is doubled. At last, the collected data is interweaved to synthesize a high resolution image. The two methods, bicubic interpolation and subpixel imaging, are simulated with Matlab7.0.1. The qualitative and quantitative analyses are given. The simulated results of lanczos and cubic spline interpolation methods are given simultaneously to show the performance of interpolation. The simulation result shows that the result images quality becomes better in both methods, but compared with bicubic interpolation method, the synthesized image PSNR gained from sub-pixel imaging is averagely enhanced by 2.1682dB, while the elapsed time is reduced to be about one forth. The subpixel imaging method could significantly mitigate the image blur introduced by under-sample.

We consider the practical NP-complete problem of finding a minimum weight spanning tree with both edge weights and inner nodes weights. We present two polynomial time algorithms with approximation factors of 2.35 · ln n and 2Hn, respectively, where n is the number of nodes in the graph and Hn is the n-th Harmonic number. This nearly matches the lower bound of (l-\in )Hn, for any \in \ge 0. We also give an approximation algorithm with approximation factor \Delta - 1, where \Delta is the maximum degree of the graph. For metric spaces, we give a 3.105-approximation algorithm and show that an approximation factor of 1.463 is impossible unless {NP \subseteq DTIME[n^{O(\log longn)} ]}.

The main feature of variable structure control systems is the sliding mode. In the sliding mode, the eigenvalues of the system are specified and the system enjoys invariance to parameter variations and disturbances. Then, how can one apply this control method to the systems with time delay? In this paper, the discrete variable structure controller is first reviewed and discussed. Combined with the predictor control method, the authors propose a variable structure predictor controller with quasi-sliding mode. Simulation results are presented illustrating the effectiveness of the control method. >