Search

Small-object detection is a basic and challenging problem in computer vision tasks. It is widely used in pedestrian detection, traffic sign detection, and other fields. This paper proposes a deep learning small-object detection method based on image super-resolution to improve the speed and accuracy of small-object detection. First, we add a feature texture transfer (FTT) module at the input end to improve the image resolution at this end as well as to remove the noise in the image. Then, in the backbone network, using the Darknet53 framework, we use dense blocks to replace residual blocks to reduce the number of network structure parameters to avoid unnecessary calculations. Then, to make full use of the features of small targets in the image, the neck uses a combination of SPPnet and PANnet to complete this part of the multi-scale feature fusion work. Finally, the problem of image background and foreground imbalance is solved by adding the foreground and background balance loss function to the YOLOv4 loss function part. The results of the experiment conducted using our self-built dataset show that the proposed method has higher accuracy and speed compared with the currently available small-target detection methods.

In modern complex industrial processes, due to poor rationalization of alarm systems and the complexity of process interconnections, alarm floods are commonly present. Alarm floods are also identified as the main causes of many industrial accidents. One valid approach to deal with alarm floods is to mine meaningful alarm sequential patterns from alarm floods. These identified patterns can help to analyze root causes or to configure dynamic alarming modules. In this paper, a method based on the combination of ClaSP and Top-K is proposed to mine interesting alarm sequential patterns from historical alarm data. Main contributions of this study are twofold: 1) A pattern mining approach is adapted to mine interesting patterns from alarm flood sequences; 2) A pattern compression strategy is proposed to reduce pattern redundancy. A case study is presented to demonstrate the effectiveness of the proposed method.

The vector nature of dissipative-soliton-resonance (DSR) pulses is numerically investigated in an all-normal-dispersion Yb-doped fiber laser mode-locked by the nonlinear polarization rotation technique. We calculate the polarization ellipses of the DSR pulse throughout the cavity, which shows that the polarization state varies during the propagation. The features of pulse polarization are also displayed on the Poincaré sphere. The impact of nonlinear birefringence on the polarization evolution is analyzed. Furthermore, we find that the polarization state is not uniform across a DSR pulse at a fixed intracavity position. The flat-top part of the pulse has a nearly fixed polarization, while its leading and trailing edges feature polarization states that vary with time. The peak power ratio of the two orthogonal polarization components in the pulse is limited in the range from 4 to 31. Our simulation results offer insight into the polarization properties of DSR pulses in mode-locked fiber lasers containing polarization discrimination devices.

Radar imaging has been one of the basic functions for modern radar as well as searching, detection and tracking, especially for those ground-based space surveillance phased array radars. Due to the periodic revisiting characteristics of the in-orbit objects, multi-pass observations for space targets (e.g. satellites) within small pitching angle variation can be accumulated, which lead to the 3D inverse synthetic aperture radar (ISAR) image formation. Consequently, the strategy of allocating the limited radar observation resource for a reliable 3D ISAR imaging result is becoming a fundamental but critical issue. In this letter, a sparse aperture based radar observation resource allocating method is proposed to facilitate the space target 3D ISAR imaging. According to the sparsity estimation of the target along the azimuth dimension, the required observation resource is calculated and assigned in accordance with the imaging task priority. Furthermore, the experimental boundary of the imaging performance with different signal-to-noise ratios (SNR) and pulse extraction rates are discussed in detail, which gives an intuitive guidance for the radar observation allocation under different application scenarios. The experimental results performed on simulated data validate the feasibility and effectiveness of the proposed method.

Sentiment analysis (SA) is to predict the sentiment polarity towards a given document. Our work focuses on sentiment analysis towards comments of legal cases, which reflects people’s concern about public safety and the impartiality of judgment result. The existing methods treat this problem as a coarse document-level classification and ignore topics information, which cannot obtain an accurate sentiment polarity towards a sentence. With topics information attended, we present a novel segment-level model, to boost the performance. Firstly, we divide a sentence into several segments delimited by some important conjunctions. We find that opinion in a segment is usually explicit. Secondly, we apply two-step attention to obtain accurate representation. To enhance the interactions between comments and topics, we apply attention mechanisms twice, once between topics and context in the segment level, and the other between segments. Thirdly, we concatenate the segment and context representation to receive final representation. It unifies two types of information into a classifier. Experimental results on real-world datasets demonstrate that our model outperforms existing methods.

Recently, the Supernumerary Robotic Limbs (SRL) have emerged in the field of robotics for enhancing the user's ability without replacing natural limbs. Lots of researches have contributed to the mechanical design and control of SRL. However, building an artificial sensory nerve pathway of SRL has been largely ignored. In this paper, we built a multi-modal sensory feedback pathway for SRL. Three kinds of sensory pathway contained 1) fingertip pressure, 2) slippage and 3) proprioception information are built under the consideration of SRL platform through electrical stimulation. According to the experiment results of event-related potential (ERP) amplitude, establish the relationship between the physical parameters of the stimulus and the sensation. In addition, enhancement potentials of human sensation based on the brain-computer interface (BCI) is found in two aspects contained: 1) improvement of subjective perception resolution; 2) Shorten the time of sensory recognition. This paper can be used to objectively benchmark the human subjective and individualized sensation, and guide researchers to build the SRL sensory feedback pathways.

In order to avoid conflicts in the air traffic scenario, this paper proposed a collision avoidance method based on deep deterministic policy gradient (DDPG). Before the entering aircraft reaches the boundary of the sector, a proper heading angle is obtained by using the proposed algorithm to guarantee a conflict free condition for all aircraft. The performance of the proposed algorithm is verified by numerical simulations. The results of the numerical simulations show that the entering aircraft is successfully guided in a sector to avoid conflicts and reaches the destination with the shortest distance, furthermore, the proposed algorithm takes less time.

Asymmetric soliton explosions are numerically demonstrated in an all-normal-dispersion Yb-doped mode-locked fiber laser based on a lumped model. It is found that the explosion events manifest themselves when the pump strength is tuned to reside between stationary dissipative soliton and noise-like pulse regimes. Remarkably, although the initial conditions and the model itself are symmetric, the explosions occur only at one edge of the soliton, i.e., the pulse leading and trailing edges explode alternately. Our results pave the way for further investigation on the dynamics of soliton explosions in ultrafast fiber lasers.

Recent advances in deep reinforcement learning have shown promising potential results in solving complex control problems. Inspired by these achievements, in this paper, an environment which can be applied in agent training to solve aircraft landing guidance problem is proposed. By this environment, the agent receives the current state of the aircraft and the runway in a sector and outputs an action to guide the aircraft to land at the runway. The Deep Q Network (DQN) algorithm is used to verify the feasibility of the environment. The experimental results show that this environment can be used to train an agent, and the guidance action produced by the agent is consistent with the behavior of air traffic controllers.

With the development of mobile communication technologies, the application of 5G key technologies such as large-scale antennas, full-spectrum access, and high-frequency millimeter waves will drive the RF devices to be synchronously upgraded, and the packaging of RF devices is getting smaller and smaller. How to accurately test RF devices Performance parameters become more and more difficult. It must design test fixtures for RF devices to accurately test the performance parameters of RF devices. This article uses the powerful modeling and computational capabilities of HFSS (High Frequency Structure Simulator) to design non-coaxial microwave test fixtures and calibration devices. To study the EMC characteristics of the fixture and to optimize the fixture structure, the TRL (Thru-Reflect-Line) calibration method is used to correct the system error introduced by the test fixture during testing of the microwave device so as to obtain the accurate S parameter of the microwave device. At last, the TRL calibration algorithm is compiled by Matlab to test the verification pieces, through the verification test pieces show that, the S11 in the calibration result is small, both below −20 dB, indicating that there is almost no signal reflection; S21 is close to 0, indicating that there is almost no signal insertion loss. The designed test fixtures and fixture calibration pieces allow accurate and rapid testing of non-coaxial RF devices.

Scattering center models of canonical structures such as flat plates and dihedral play an important role in automatic target recognition. Based on the far-field radiation integral formula, the analytic bistatic 3D parametric scattering center model of the spatial separate plates is derived in detail. Taking the plate spacing variable into account, the geometrical optics(GO) method and the physical optics(PO) method are combined to analyze the influence of different angles of incidence and target size on the illuminated area. The total scattering solutions of spatial separate plates composed of first- and second-order scattering are presented in the form of horizontal and vertical polarization. Compared with the SBR algorithm of electromagnetic calculation software CST, the modeling results greatly improve the computational efficiency while maintaining the precision of the simulated echo, which verifies the validity of the proposed model.

The tropospheric scatter mechanism is potentially valuable for over-the-horizon (OTH) passive sensing, two limitations of this mechanism as large propagation losses and scattering fading diffusion are analyzed. Based on the diffusion properties, the received signals are analyzed with a bilinear Cohen's class distribution, of which the parameterization function (i.e. the kernel) is built from the diffusion coefficient to enhance the SNR. After that, target signals are focused in a parameter space by an equal-slope Hough transform (HT) with a priori information of the intra-pulse frequency interval, the focus-points can be effectively accumulated to detect the OTH targets.

The Over-The-Horizon (OTH) locating of radiation source can be realized by the troposcatter propagation of electromagnetic (EM) signals. This paper studies the troposcatter signal group delay for target ranging and proposes a modified beamspace MUSIC to estimate the arrival azimuth angle of the troposcatter signal. Finally, the OTH target can be located by the ranging and azimuth angle estimation results.

Space target ISAR image components segmentation is an important foundation of space target recognition based on ISAR image. Considering the difference between the ISAR image and optical image and the main body has the diverse distribution of pixels from the subcomponent in the ISAR image, this paper proposes a segmentation idea that subcomponent segments after the main body based on CLEAN: firstly, we develop an adaptive region growing method to segment the main body that emphatically analyses and solves the setting problem of the threshold of region growing method; then, in order to divide the residual subcomponent, we exploit the proximity of pixels in the ISAR image to propose an iterative correlation clustering method based on perceptual organization. Finally, some simulation experiments are conducted to verify the validity of our methods.

This letter presents a novel 3-D coherent ISAR imaging algorithm based on mono-static radar observation. Due to the intrinsic orbit characteristics of the space target, the pitch dimension resolution is obtained by utilizing multiple orbits observation relative to radar. For the sparse characteristic of the target echo signal, the information of all resolution cells from the same image is also taken into account under the condition of sparse constraint to achieve coherence processing. Compared with the directly Fourier transform algorithm, the proposed method can effectively suppress the false scattering point target and achieve more accurate 3-D imaging result.

A new indoor calibration method is proposed to estimate and compensate mutual coupling and channel gain/phase inconsistency parameters for a uniform linear array (ULA). This calibration method employs only one assistant source in near-field and is unnecessary to conduct complex feed circuits behind the antennas. Firstly, accurate channel gain/phase deviation is estimated via an inside testing structure exploiting the coherence of a single-frequency test signal. Then, the received array manifold of the near-field auxiliary source is modeled with a second-order Taylor series approximate. Finally, joint mutual coupling and gain/phase error are compensated using the inside testing results and modeled array manifold as the initialization of the eigenstructure (also called subspace) technique, which profoundly accelerates the convergence and guarantees the uniqueness of the converged solution. Experiments conducted on an actual 16-element ULA validate the proposed method.

Microwave scattered by the troposphere can propagate Over-The-Horizon(OTH), how to effectively detect the received signal using its signal characteristics is the most important issue in troposcatter target reconnaissance. The received signal behaves with low signal to noise ratio (SNR), multipath fading. In order to effectively detect, the SNR needs to be enhanced utilizing the signal characteristics. This paper analyzes the signal multipath fading and finds energy diffusion characteristics affected by multipath fading in the time and frequency domain. Based on this, an appropriate signal truncation strategy is designed respectively in time, frequency domain to obtain the energy-concentrated part to maximize truncation gain. Next, the signal detection is carried out after the paper accumulates signal energy of the truncated part. The algorithm does not need any prior information which is effective in the non-cooperative circumstance. Finally, the simulation results validate the detection performance of the proposed algorithm is better than the traditional energy detection. The proposed one can satisfy the detection need of the troposcatter signal.

Today's datacenter is shared among various applications with different QoS requirements, which poses a great challenge to deliver low delay transport with high throughput. Most of works address this challenge by reducing the in-network delay, but assumes a negligible local delay. However, we show that this assumption does not hold for a multi-tenant datacenter that a physical machine is shared by multiple tenants with virtual machines running different applications. As measured, we found that VMs in a PM competing for bandwidth resources introduce delays as high as 13 ms, resulted from the packet queueing at QDisc layer of that PM, because current VMs' rate control still operates in a distributed manner without exploiting knowledge of the QoS requirements of applications running in VMs. This work addresses this problem by proposing a centralized rate adaptation (CERA) that operates in the host PM, dynamically schedules the flows from all VMs in a centralized manner. We implemented a CERA prototype and evaluated CERA through testbed experiments. Our results show that CERA reduces the local delay significantly thus reduces the average request latency of delay sensitive applications, e.g., memcached, by a factor of 6.3, without sacrificing the throughput performance of throughput intensive applications, e.g., iperf.

A 4-node quadrilateral piezoelectric element with thin piezoelectric layers embedded on top and bottom surfaces is derived based on the first-order shear deformation theory, and the dynamic equation of a composite beam with distributed sensors and actuators is established by Hamilton's principle. Considering the influence of plant noise and measurement noise, linear quadratic Gauss (LQG) control algorithm is obtained by using Kalman filter for state estimation. The active vibration control of a piezoelectric composite beam subjected to the pulse, harmonic and random excitation is achieved by LQR and LQG control algorithm, respectively. The results indicate that, if the effect of the noise is taken into consideration and the entire states cannot be fully observed, the LQG algorithm demands a lower actuation voltage than LQR does.

DOA estimation is a hot topic in tropospheric scatter radar target reconnaissance Over-The-Horizon (OTH). The received signal propagated through troposphere OTH is so weak because of strong loss, the classical MUSIC has been unable to effectively estimate angle. So we adopt Time-Frequency (TF) MUSIC, i.e., replacing the array covariance matrix of the classical MUSIC with Time-Frequency Distribution (TFD) matrix, which is suitable for the low Signal-to-Noise Ratio (SNR) condition. In order to effectively processing signal, the troposcatter signal characteristics should be modeled and analyzed. Firstly, we analyze troposcatter propagation loss, and further deduce the receiver output SNR to provide support for algorithm design. Secondly, considering that time, frequency diffusion in energy caused by troposcatter multipath fading, we derive time-delay and frequency-shift diffusion functions based on scatter transmission function. To be more specific, the diffusion characteristics are studied on the variations of the beam width, target distances etc.. Then we deduce the joint time-frequency diffusion function and coefficient from the above two diffusions. Affected by the diffusion, the TFD matrix will be distorted to degrade the algorithm accuracy. To utilize the characteristic, we propose a weighted scattering signal TFD matrix, where the weight matrix is derived from the joint diffusion coefficient, to prominent the scattering signal according to the matched filter theory. Then we can estimate exactly angle for the troposcatter signal using the weighted matrix.

The optical properties of Si nanocone structures were theoretically studied by simulations. Fig. 1(a) illustrates the configuration of nanocone arrays used in the simulations. The nanocone arrays in our simulations is defined by three parameters: base diameter BD, period P and height H, and the base diameter is always set equal to the period. Simulations were performed to compare the normalized scattering cross section (Q sca ) of an isolated nanocone and that of a compact nanocone array. As shown in Fig. 1(b), the compact periodic structure strongly affects the modes of Mie resonance. The resonance modes merge into a broader and much stronger feature at 500 to 1000nm range, which is in agreement with previous works on nanoparticles [1]. In Fig. 1(c), we can see that the reflectance reaches to a minimal value when Q sca is maximal, which clearly shows that Mie resonace modes play an important role in high absorbance of nanocone structure. The electric field distribution in Fig. 1(d) demonstrates how light was coupled into the substrate by Mie resonance modes. The steep descend corresponding to 413nm in Fig. 1(c) is due to Wood-Rayleigh (WR) anomaly effect caused by the period structure [2].

The steadiness of the orbit is crucial for the performance of the space target 3D reconstruction. The relationship between the orbit characteristics and the steadiness of the target moving trajectory is investigated in this paper, and the target 3D reconstruction performance for both steady and unsteady moving trajectories are quantitatively analyzed through the simulation experiments.

Stochastic model predictive control (SMPC) can be used in a broad variety of fields, such as diverse optimization problems. One issue which recently emerges is that some random variables contained in the objective function or probabilistic constraints will have compound distributions. The compound distributions are determined by some parameters which themselves are random and follow certain known distributions. This paper will propose an effective and intelligible way to compute compound distributions with the use of a similar logic to the Falling Shadow Theory. And to tackle the stochastic objective function, two typical formulations in risk studies, i.e., the value-at-risk (VaR) and conditional value-at-risk (CVaR), are introduced and formulated for SMPC. A simulation example regarding a coal trade decision-making process shows the efficacy of the proposed strategy.

Mobile data networks exhibit highly variable data rates and stochastic non-congestion-related packet loss. These challenges result in key performance bottlenecks in current Transmission Control Protocol (TCP) implementations: bandwidth inefficiency and large end-to-end delay. This work addresses these challenges by first developing a Sliding Interval based Rate Adaptation (SIRA) that tracks bandwidths with a fixed time interval and applies them to its transmission rate periodically. Extensive experiments confirmed that SIRA achieves 96.3% bandwidth utilization and reduces the average queueing delay by a factor of 1.37, compared to TCP CUBIC, the preferred variant for Internet servers. However, the resultant end-to-end delay is still much larger for interactive applications, thus we complement SIRA with two heuristic rate compensation algorithms (SIRA-H) given that the bandwidth does not vary significantly in long time scales. Specifically, SIRA-H first reduces the transmission rate of SIRA if the estimated RTT is above a prefigured threshold. Meanwhile, it computes the amount of unsent data that would be transmitted if SIRA were used, and compensates the rate reduction with those unsent data as if their ACKs were received, when the queue is detected to be empty. We evaluated SIRA-H through a combination of trace-driven emulations and real-world experiments, and showed that it reduces the 95th percentile queueing delay by a factor of over 3.9, while maintains a similar throughput compared to the original SIRA. In comparison to state of the art protocols such as Sprout and Verus, SIRA-H also reduces the 95th percentile queueing delay by a factor of over 0.8.

Troposcatter array signal model and processing algorithm are of great importance when over-the-horizon passive reconnaissance is carried out with troposcatter. This paper studies on troposcatter array signal model, and the diffusion phase between different array elements is found after analyzing the inter-relationship of different array element transfer functions,. Moreover, when analyzing the diffusion phase change of different targets, it is known that the bigger the target azimuth angle is, the further the distance between the transmitting and receiving antenna is, the wider the target azimuth beam width is, the bigger the array element number is, then the more obvious the diffusion phase is. Finally, aiming at the characteristics of troposcatter signal model, this paper improves the MUSIC algorithm, and designs the array signal processing algorithm which can estimate the target spatial distance spectrum.

Because of the positive and inverse piezoelectric effect, piezoelectric materials can be used in the active control of flexible structures. Based on the first-order shear deformation theory, combined with the electrical and mechanical effects, a 4-node quadrilateral piezoelectric composite element is derived, and the dynamic characteristic equation of a composite beam is established by Hamilton's principle. Considering the influence of outside noise and measurement noise, using Kalman filter for state estimation, the linear quadratic Gauss (LQG) control algorithm is obtained based on LQR algorithm. The vibration control of a piezoelectric beam in the pulse, harmonic and random excitation is achieved by LQR and LQG control algorithm, respectively. The results indicate that, without considering the entire states and the impact of noise as prerequisites, the LQG algorithm needs a smaller actuation voltage than LQR does.

The troposphere scatter mechanism is potential for the passive detection of various over-the-horizon (OTH) targets such as radar, in which the prediction of propagation loss plays an important role. According to the latest recommendation of Radiocommunication Sector of International Telecommunication Union (ITU-R), this paper establishes a model to effectively predict the propagation loss for passive detection, which describes the dependence of propagation loss with respect to several important factors including the receiving/transmitting beam angles, the target distance and the height of receiving antenna. Furthermore, the detectable region of the passive receiver is also derived for the OTH targets. Simulation result shows the effectiveness of the model and the theoretical analysis.

Traffic speed is one of the most important quantities for travel information systems. Accurate speed forecasting can help in trip planning by allowing travelers to avoid the congested routes, either by choosing the alternative routes or by changing the departure time. It is also helpful for traffic monitoring, control, and planning. An important feature of traffic is that it consists of free flow and congested regimes, which have significantly different properties. Training a single traffic speed predictor for both regimes typically results in suboptimal accuracy. To address this problem, a mixture of experts algorithm which consists of two regime-specific linear predictors and a decision tree gating function was developed. A generalized expectation maximization algorithm was used to train the linear predictors and the decision tree. The proposed algorithm was evaluated on a 5-mile stretch of I35 highway in Minneapolis containing 10 single loop detector stations, with prediction horizons ranging from 5 minutes to one hour ahead. Experimental results showed that mixture of experts approach outperforms several popular benchmark approaches.

Height estimation of man-made structures in urban and suburban areas is an interesting but challenging problem, especially from single SAR image, which usually occurs in emergency response after crisis. A novel model based height estimation approach is proposed, which finds a best match between parameterized object model and the observed single intensity SAR image data. Firstly, a projection function for single bounce signal return between 3D object model and corresponding characteristic regions in SAR image is constructed, which make the model generation phase very simple and fast. Then, a new likelihood measure for the matching between the height hypothesis and the SAR data is proposed in the respect of image segmentation. The matching procedure is iterated until a global maximum similarity is achieved and then the height estimation is obtained, which is realized by the simulated annealing algorithm. The experiments performed on general terms of buildings in the simulated SAR image data set verify its validity.

With the improvements of spaceborne and airborne SAR system resolution, the applications of radar remote sensing has been extended to building 3D geometric information retrieval and reconstruction from urban SAR images, which is the foundation of build-up areas reconstruction and urban analysis. This paper mainly focuses on the problem of building height estimation from a single high resolution (HR) SAR image of urban scenes. A model based method combined with image segmentation framework of building height estimation is proposed. This method optimizes a new likelihood measure between the projection image from 3D geometric model of the buildings and the observed image over the heights hypothesis space. With assumption of the parallelepiped shapes, the SAR building area is partitioned into several regions. The new likelihood criterion then measures both the inner homogeneity of partitioned regions as well as their inter heterogeneity to achieve robust height hypothesis test. The optimization is done by simulated annealing in order to avoid local optimum. The experimental results performed on simulated SAR image data set valid the proposed method.

A fast online algorithm OnlineSVMR for training Ramp-Loss Support Vector Machines (SVMRs) is proposed. It finds the optimal SVMR for t+1 training examples using SVMR built on t previous examples. The algorithm retains the Karush–Kuhn–Tucker conditions on all previously observed examples. This is achieved by an SMO-style incremental learning and decremental unlearning under the Concave-Convex Procedure framework. Further speedup of training time could be achieved by dropping the requirement of optimality. A variant, called OnlineASVMR, is a greedy approach that approximately optimizes the SVMR objective function and is suitable for online active learning. The proposed algorithms were comprehensively evaluated on 9 large benchmark data sets. The results demonstrate that OnlineSVMR (1) has the similar computational cost as its offline counterpart; (2) outperforms IDSVM, its competing online algorithm that uses hinge-loss, in terms of accuracy, model sparsity and training time. The experiments on online active learning show that for a fixed number of label queries OnlineASVMR (1) achieves consistently better accuracy than QueryAll and competitive accuracy to Greedy approach; (2) outperforms the active learning version of IDSVM.

Kernel Perceptrons are represented by a subset of training points, called the support vectors, and their associated weights. To address the issue of unlimited growth in model size during training, budget kernel perceptrons maintain the fixed number of support vectors and thus achieve the constant update time and space complexity. In this paper, a new kernel perceptron algorithm for online learning on a budget is proposed. Following the idea of Tighter Perceptron, upon exceeding the budget, the algorithm removes the support vector with the minimal impact on classification accuracy. To optimize memory use, instead on maintaining a separate validation data set for accuracy estimation, the proposed algorithm only uses the support vectors for both model representation and validation. This is achieved by estimating posterior class probability of each support vector and using this information in validation. The experimental results on 11 benchmark data sets indicate that the proposed algorithm is significantly more accurate than the competing budget kernel perceptrons and that it has comparable accuracy to the resource unbounded perceptrons, including the original kernel perceptron and the Tighter Perceptron that uses whole training data set for validation.

Kernel machines are a popular class of machine learning algorithms that achieve state of the art accuracies on many real-life classification problems. Kernel perceptrons are among the most popular online kernel machines that are known to achieve high-quality classification despite their simplicity. They are represented by a set of B prototype examples, called support vectors, and their associated weights. To obtain a classification, a new example is compared to the support vectors. Both space to store a prediction model and time to provide a single classification scale as O(B). A problem with kernel perceptrons is that on noisy data the number of support vectors tends to grow without bounds with the number of training examples. To reduce the strain at computational resources, budget kernel perceptrons have been developed by upper bounding the number of support vectors. In this work, we propose a new budget algorithm that upper bounds the number of bits needed to store kernel perceptron. Setting the bitlength constraint could facilitate development of hardware and software implementations of kernel perceptrons on resource-limited devices such as microcontrollers. The proposed compressed kernel perceptron algorithm decides on the optimal tradeoff between number of support vectors and their bit precision. The algorithm was evaluated on several benchmark data sets and the results indicate that it can train highly accurate classifiers even when the available memory budget is below 1 Kbit. This promising result points to a possibility of implementing powerful learning algorithms even on the most resource-constrained computational devices.

In order to use the spaceborne optical imaging and electronic reconnaissance data to monitor the ships formation, this paper proposes a target association matching algorithm based on the structure of ships formation and targetspsila attribute. Simulation results that this algorithm delivers better performance compared to the traditional method, in the situation that because of the sensors are sparse sampling and the targets location precision are greatly different so that can not be established to creation the targetspsila motion model.

Under the frame of Dempster-Shafer theory of evidence, a distance function to depict comparability between evidences is constructed according to the conflict among evidences, which is for the case that the origin of few evidences is uncertain. In order to conquer these disadvantages of traditional quick k-nearest neighbor (k-NN) classification algorithm, this paper proposes a quick k-NN evidence classification algorithm-super-ball search evidence classification (ab. S-BSEC) algorithm based on near neighbor searching. Simulation results show that this method is superior to the traditional k-NN algorithm in terms of the recognition speed under the same recognition rate and k, and super-ball algorithm is not sensitive to searching order of training sample.

A realization of fuzzy logic by a neural network is described. Each node in the network represents a premise or a conclusion. Let x be a member of the universal set, and let A be a node in the network. The value of activation of node A is taken to be the value of the membership function at point x, m/sub A/(x). A logical operation is defined by a set of weights which are independent of x. Given any value of x, a preprocessor will determine the values of the membership function for all the premises that correspond to the input nodes. These are treated as input to the network. A propagation algorithm is used to emulate the inference process. When the network stabilizes, the value of activation at an output node represents the value of the membership function that indicates the degree to which the given conclusion is true. Weight assignment for the standard logical operations is discussed. It is also shown that the scheme makes it possible to define more general logical operations. >

The authors discuss the design of a rule-adaptive fuzzy control system, and demonstrate its usefulness in controlling plants of various dynamical natures. The rule-adaptive fuzzy controller generates the fuzzy control signal as a convex combination of the standard fuzzy inputs to the controller, the error, and the error rate. Such a combination is automatically adjusted online amid the varying control situations with a certain updating scheme. Two different updating schemes were applied to the simulations of four selected dynamical plants. The simulation results show that the controller is effective in controlling a wide range of dynamical plants. >

A unified mathematical approach for the study of the theory of inference is presented. The authors introduce the notion of inference channels and consider the truth values flowing among these channels. They study the mathematical structure of these inference channels based on a framework of knowledge representation and thus establish a theory of inference to gain an insight into the core of approximate reasoning. >