This paper concerns the problem of object segmentation in real-time for picking system. A region proposal method inspired by human glance based on the convolutional neural network is proposed to select promising regions, allowing more processing is reserved only for these regions. The speed of object segmentation is significantly improved by the region proposal method. By the combination of the region proposal method based on the convolutional neural network and superpixel method, the category and location information can be used to segment objects and image redundancy is significantly reduced. The processing time is reduced considerably by this to achieve the real time. Experiments show that the proposed method can segment the interested target object in real time on an ordinary laptop.
Unmanned aerial vehicles (UAVs) may play an important role in data collection and offloading in vast areas deploying wireless sensor networks, and the UAV's action strategy has a vital influence on achieving applicability and computational complexity. Dynamic programming (DP) has a good application in the path planning of UAV, but there are problems in the applicability of special terrain environment and the complexity of the algorithm. Based on the analysis of DP, this paper proposes a hierarchical directional DP (DDP) algorithm based on direction determination and hierarchical model. We compare our methods with Q-learning and DP algorithm by experiments, and the results show that our method can improve the terrain applicability, meanwhile greatly reduce the computational complexity.
To overcome the shortcomings of the Lee image enhancement algorithm and its improvement based on the logarithmic image processing (LIP) model, this paper proposes what we believe to be an effective image enhancement algorithm. This algorithm introduces fuzzy entropy, makes full use of neighborhood information, fuzzy information and human visual characteristics. To enhance an image, this paper first carries out the reasonable fuzzy-3 partition of its histogram into the dark region, intermediate region and bright region. It then extracts the statistical characteristics of the three regions and adaptively selects the parameter α according to the statistical characteristics of the image’s gray-scale values. It also adds a useful nonlinear transform, thus increasing the ubiquity of the algorithm. Finally, the causes for the gray-scale value overcorrection that occurs in the traditional image enhancement algorithms are analyzed and their solutions are proposed. The simulation results show that our image enhancement algorithm can effectively suppress the noise of an image, enhance its contrast and visual effect, sharpen its edge and adjust its dynamic range.
Accurate prediction of server load is important to cloud systems for improving the resource utilization, reducing the energy consumption and guaranteeing the quality of service (QoS). This paper analyzes the features of cloud server load and the advantages and disadvantages of typical server load prediction algorithms, integrates the cloud model (CM) and the Markov chain (MC) together to realize a new CM-MC algorithm, and then proposes a new server load prediction algorithm based on CM-MC for cloud systems. The algorithm utilizes the historical data sample training method of the cloud model, and utilizes the Markov prediction theory to obtain the membership degree vector, based on which the weighted sum of the predicted values is used for the cloud model. The experiments show that the proposed prediction algorithm has higher prediction accuracy than other typical server load prediction algorithms, especially if the data has significant volatility. The proposed server load prediction algorithm based on CM-MC is suitable for cloud systems, and can help to reduce the energy consumption of cloud data centers.
The particle filter (PF) algorithm is one of the most commonly used algorithms for maneuvering target tracking. The traditional PF maps from multi-dimensional information to onedimensional information during particle weight calculation, and the incorrect transmission of information leads to the fact that the particle prediction information does not match the weight information, and its essence is the reduction of the information entropy of the useful information. To solve this problem, a dual channel independent filtering method is proposed based on the idea of equalization mapping. Firstly, the particle prediction performance is described by particle manipulations of different dimensions, and the accuracy of particle prediction is improved. The improvement of particle degradation of this algorithm is analyzed in the aspects of particle weight and effective particle number. Secondly, according to the problem of lack of particle samples, the new particles are generated based on the filtering results, and the particle diversity is increased. Finally, the introduction of the graphics processing unit (GPU) parallel computing the platform, the "channel-level" and "particlelevel" parallel computing the program are designed to accelerate the algorithm. The simulation results show that the algorithm has the advantages of better filtering precision, higher particle efficiency and faster calculation speed compared with the traditional algorithm of the CPU platform.
In order to achieve a high precision in three-dimensional (3D) multi-camera measurement system, an efficient multi-camera calibration method is proposed. A stitching method of large scale calibration targets is deduced, and a fundamental of multi-camera calibration based on the large scale calibration target is provided. To avoid the shortcomings of the method, the vector differences of reprojection error with the presence of the constraint condition of the constant rigid body transformation is modelled, and minimized by the Levenberg-Marquardt (LM) method. Results of the simulation and observation data calibration experiment show that the accuracy of the system calibrated by the proposed method reaches 2 mm when measuring distance section of 20 000 mm and scale section of 7 000 mm × 7 000 mm. Consequently, the proposed method of multi-camera calibration performs better than the fundamental in stability. This technique offers a more uniform error distribution for measuring large scale space.
The disturbance torque of aerostatic bearings is in the same order of the reaction wheel, which causes difficulty in evaluation of the designed attitude control strategy of a nano-satellite based on the aerostatic bearing. Two approaches are proposed to model the disturbance torque. Firstly, the gravity induced moment, the vortex torque, and the damping moment are modeled separately. However, the vortex torque and the damping moment are coupled with each other as both of them are caused by the viscosity. In the second approach, the coupling effect is considered. A nano-satellite is constructed based on aerostatic bearing. The time history of the free rotation rate from an initial speed is measured by the gyro, which is further used to calculate the rotation angle and acceleration. The static vortex torque is measured via the removable micro-torque measurement system. Based on these data, the model parameters are identified and modeling errors are presented. Results show that the second model is more precise. The root mean squire error is less than 0.5×10-4 N·m and the relative error of the static vortex torque is 0.16%.
The time dependent vehicle routing problem with time windows (TDVRPTW) is considered. A multi-type ant system (MTAS) algorithm hybridized with the ant colony system (ACS) and the max-min ant system (MMAS) algorithms is proposed. This combination absorbs the merits of the two algorithms in solutions construction and optimization separately. In order to improve the efficiency of the insertion procedure, a nearest neighbor selection (NNS) mechanism, an insertion local search procedure and a local optimization procedure are specified in detail. And in order to find a balance between good scouting performance and fast convergence rate, an adaptive pheromone updating strategy is proposed in the MTAS. Computational results confirm the MTAS algorithm's good performance with all these strategies on classic vehicle routing problem with time windows (VRPTW) benchmark instances and the TDVRPTW instances, and some better results especially for the number of vehicles and travel times of the best solutions are obtained in comparison with the previous research.
The rapid development of data communication in modern era demands secure exchange of information. Steganography is an established method for hiding secret data from an unauthorized access into a cover object in such a way that it is invisible to human eyes. The cover object can be image, text, audio, or video. This paper proposes a secure steganography algorithm that hides a bitstream of the secret text into the least significant bits (LSBs) of the approximation coefficients of the integer wavelet transform (IWT) of grayscale images as well as each component of color images to form stego-images. The embedding and extracting phases of the proposed steganography algorithms are performed using the MATLAB software. Invisibility, payload capacity, and security in terms of peak signal to noise ratio (PSNR) and robustness are the key challenges to steganography. The statistical distortion between the cover images and the stego-images is measured by using the mean square error (MSE) and the PSNR, while the degree of closeness between them is evaluated using the normalized cross correlation (NCC). The experimental results show that, the proposed algorithms can hide the secret text with a large payload capacity with a high level of security and a higher invisibility. Furthermore, the proposed technique is computationally efficient and better results for both PSNR and NCC are achieved compared with the previous algorithms.
Fusion methods based on multi-scale transforms have become the mainstream of the pixel-level image fusion. However, most of these methods cannot fully exploit spatial domain information of source images, which lead to the degradation of image. This paper presents a fusion framework based on block-matching and 3D (BM3D) multi-scale transform. The algorithm first divides the image into different blocks and groups these 2D image blocks into 3D arrays by their similarity. Then it uses a 3D transform which consists of a 2D multi-scale and a 1D transform to transfer the arrays into transform coefficients, and then the obtained low- and high- coefficients are fused by different fusion rules. The final fused image is obtained from a series of fused 3D image block groups after the inverse transform by using an aggregation process. In the experimental part, we comparatively analyze some existing algorithms and the using of different transforms, e.g. non-subsampled Contourlet transform (NSCT), non-subsampled Shearlet transform (NSST), in the 3D transform step. Experimental results show that the proposed fusion framework can not only improve subjective visual effect, but also obtain better objective evaluation criteria than state-of-the-art methods.
Artificial bee colony (ABC) is one of the most popular swarm intelligence optimization algorithms which have been widely used in numerical optimization and engineering applications. However, there are still deficiencies in ABC regarding its local search ability and global search efficiency. Aiming at these deficiencies, an ABC variant named hybrid ABC (HABC) algorithm is proposed. Firstly, the variable neighborhood search factor is added to the solution search equation, which can enhance the local search ability and increase the population diversity. Secondly, inspired by the neuroscience investigation of real honeybees, the memory mechanism is put forward, which assumes the artificial bees can remember their past successful experiences and further guide the subsequent foraging behavior. The proposed memory mechanism is used to improve the global search efficiency. Finally, the results of comparison on a set of ten benchmark functions demonstrate the superiority of HABC.
This paper introduces an optimization algorithm, the hummingbirds optimization algorithm (HOA), which is inspired by the foraging process of hummingbirds. The proposed algorithm includes two phases: a self-searching phase and a guide-searching phase. With these two phases, the exploration and exploitation abilities of the algorithm can be balanced. Both the constrained and unconstrained benchmark functions are employed to test the performance of HOA. Ten classic benchmark functions are considered as unconstrained benchmark functions. Meanwhile, two engineering design optimization problems are employed as constrained benchmark functions. The results of these experiments demonstrate HOA is efficient and capable of global optimization.
Face recognition based on few training samples is a challenging task. In daily applications, sufficient training samples may not be obtained and most of the gained training samples are in various illuminations and poses. Non-sufficient training samples could not effectively express various facial conditions, so the improvement of the face recognition rate under the non-sufficient training samples condition becomes a laborious mission. In our work, the facial pose pre-recognition (FPPR) model and the dualdictionary sparse representation classification (DD-SRC) are proposed for face recognition. The FPPR model is based on the facial geometric characteristic and machine learning, dividing a testing sample into full-face and profile. Different poses in a single dictionary are influenced by each other, which leads to a low face recognition rate. The DD-SRC contains two dictionaries, full-face dictionary and profile dictionary, and is able to reduce the interference. After FPPR, the sample is processed by the DD-SRC to find the most similar one in training samples. The experimental results show the performance of the proposed algorithm on olivetti research laboratory (ORL) and face recognition technology (FERET) databases, and also reflect comparisons with SRC, linear regression classification (LRC), and two-phase test sample sparse representation (TPTSSR).
Dempster-Shafer evidence theory (DS theory) is widely used in brain magnetic resonance imaging (MRI) segmentation, due to its efficient combination of the evidence from different sources. In this paper, an improved MRI segmentation method, which is based on fuzzy c-means (FCM) and DS theory, is proposed. Firstly, the average fusion method is used to reduce the uncertainty and the conflict information in the pictures. Then, the neighborhood information and the different influences of spatial location of neighborhood pixels are taken into consideration to handle the spatial information. Finally, the segmentation and the sensor data fusion are achieved by using the DS theory. The simulated images and the MRI images illustrate that our proposed method is more effective in image segmentation.
An algorithm is presented for image prior combinations based blind deconvolution and applied to astronomical images. Using a hierarchical Bayesian framework, the unknown original image and all required algorithmic parameters are estimated simultaneously. Through utilization of variational Bayesian analysis, approximations of the posterior distributions on each unknown are obtained by minimizing the Kullback-Leibler (KL) distance, thus providing uncertainties of the estimates during the restoration process. Experimental results on both synthetic images and real astronomical images demonstrate that the proposed approaches compare favorably to other state-of-the-art reconstruction methods.
