Search

Few-shot learning (FSL) inspired by deep neural network and its practice on electromagnetic susceptibility evaluation are reported. By the assistance of model, metric or optimization, FSL can not only make use of the prior information, but also gain the ability of learning the pattern for distinguishing and classifying input vectors from only a few samples. For transient electromagnetic disturbances, the recurrent neural network (RNN) is involved to help with the memory of the input signal with the style of 1-D time sequence. In order to learn the proper feature of transient disturbances, a Siamese neural network containing two RNNs is built and trained, and the distance of two outputs can be measured by a preselected distance function. This makes the proposed model capable to learn the difference between normal cases and failure cases. In order to demonstrate this method, a pulse current injection testing on a certain electronic system is conducted and evaluated with the metric-based FSL. The result confirms the practicability of the proposed model, for the correct classification rate can be 83.3% with only 36 samples.

In order to solve the current situation of low management efficiency in university laboratories and the inconvenience to use it, this paper proposes and designs an intelligent experiment management system based on Internet of things technology. This system puts forward a feasible and detailed design scheme, which adopts the Internet of things technology to design the intelligent experiment management system and uses Java to realize the intelligent experiment management platform. Finally, the function of the management system is tested and verified. The test shows that the intelligent experiment management system has basically achieved the expected functions and meets the needs of the intelligent experiment management platform of universities.

High-power electromagnetic (HPEM) environments are capable of causing effects on systems, like malfunctions, performance degradation, interference and even collapse. To research susceptibility of systems under HPEM, experiment is believed as an efficient approach. But for some valuable and complex systems, we have no replacement nor technology to recover faulted system; otherwise, it will cost a heavy price. So we can't perform a lot of experiments and it's hard to collect enough effect data. How to assess susceptibility of systems under HPEM with limited effect data becomes an urgent problem for us. In this paper we introduce a novel statistical method, autoregressive co-kriging model, to deal with this problem. In this model, the real but small experimental data can be merged with rough but big data, such as related state experimental data and experience-based judgment data. Meanwhile, Gaussian process classification (GPC) is used to transform experimental binary data into continuous probability data to fit autoregressive co-kriging model. At the end of this paper, a case study of the susceptibility assessment of a supervisory control and data acquisition (SCADA) system under electromagnetic pulse is illustrated.

The Marx generator is an important device in pulse power technology for generating high voltage pulses. In general, the equivalent circuit of Marx generator circuit is a lumped parameter circuit under the condition of electrically small. However, for a Marx generator generating very short pulses such as ultra-wideband pulses, it is necessary to consider the wave process of pulse forming from the original terminal to the output terminal. In this paper, the circuit model of Marx generator involved distributed transmission lines is introduced, which considering the wave process. The difference between the newly raised Marx generator circuit model and the traditional Marx circuit is demonstrated by simulation and experiment. Based on this circuit model, it is useful to explain the pulse forming process accurately and to optimize the pulser design.

According to the requirements of the clean-up locomotive in the industrial production process, the locomotive in rail need to automatic complete "moving, locating, parking" reciprocating operation. Accurate locating is realized in the case of no-man control and operation safely. The principle of the locating system is to use the encoder installed on the locomotive to count for the rack fixed on the rail. The pulse number of encoder count is compared with locating parameter which was originally set, and the locomotive is adjusted and located to the accurate position. The discontinuous racks are installed on the rail at the corresponding pithead to solve the absolute accuracy of locating. The minimum locating error can be 2mm, and the maximum is 6mm when debugging on a rail up to 30000mm. This locating error meets the requirement of the system permitting error 20mm.

Creativity, one of the important abilities of industrial designers, can help designing and realizing innovative products. Today industrial design requires group collaboration too. How to effectively stimulate group creativity is an important issue on promoting innovation on products. Both the 635 brainstorming method and the C-Sketch method realize team-based divergent thinking techniques. The former uses words while the latter uses pictures as a way to stimulate group creativity. This study fuses the 635 brainstorming method and the C-Sketch method in an attempt to help group designers, by alternatively passing pictures and words in the divergent thinking phase, stimulating and integrating their ideas from different aspects for developing innovative products. To evaluate the feasibility of this hybrid technique, this study conducts two experiments. The results in the first experiment, using the proposed method, the originality and elaboration of the group creativity are significantly better than using the 635 brainstorming method, but the fluency and flexibility are not. In the second experiment, using the proposed method, the originality and elaboration of the group creativity are significantly better than using the C-Sketch method, but the flexibility is not. The C-Sketch method performs significantly better, nevertheless, than the proposed method in the fluency of group creativity. The proposed method is significantly better than the 635 brainstorming method and the C-Sketch method in improving designers' creativity thinking. Nevertheless, the measurements on the performances on the final products yield no significant differences between the proposed method and the 635 brainstorming method or the C-Sketch method.

There has been many researches[1] on EMP effects caused by kinds of incident waveforms, and statistical methods were involved on very early time to assist the analysis of effect data and results. Instand of studying the whole process and mechanics directly, this paper built up a black-box model for EMP effects evaluation based on threshold probability and logistic regression. Experiments had been done to get the effect data, with which the incident waveform characteristics were drawn and compared. The result showed that several characteristics had more correlation with effect results than others.

This paper investigates the convergence and error evaluation of the distributed analytical representation and iterative technique (DARIT-field) which is a new approach for the analysis of field coupling to multiconductor transmission lines. The DARIT-field method has the advantage of high computational efficiency over the other methods. In order to know the convergence speed and the error at each iteration step of DARIT-field, an analytic expression of iterative error is derived by combining the two telegrapher's equations into a matrix equation and using the euclidean norm to explore its upper bound. The expression shows that the convergence speed is mainly influenced by three parameters, namely coupling factor (CF), terminal loads, and the line length to excitation field wavelength ratio (d/λ). The convergence speed is a function of CF, terminal loads, and the line length to excitation field wavelength ratio (d/λ). These results allow the users to make a compromise between computational cost and accuracy by selecting the number of iterations

In this paper, we propose a new Steady State Visual Evoked Potential (SSVEP)-based brain-computer interface (BeI) with the visual stimuli displayed on the windshield via a Head Up Display system. Two chessboard images flickering at specific frequencies and representing different commands were developed as SSVEP stimuli. A linear discriminant analysis classifier was used to recognize the different patterns and output the corresponding commands. Experimental results from three healthy participants with no subject screening being conducted indicate that the average accuracy of the new BeI system is close to 80% with the window length of 4 seconds and step size of 0.25 second, showing that the proposed BeI system is feasible. The potential benefit of this system is that it little obstructs users to look at surroundings while applied to drive a vehicle.

In this paper, to develop a hybrid brain-computer interface based on Steady State Visual Evoked Potential(SSVEP) and P300, we validate a SSVEP brain-computer interface (BCI) with the hybrid stimuli of SSVEP and P300, and analyze the effect of P300 visual stimuli on the classification accuracy of the SSVEP BCI. The hybrid stimuli include the SSVEP stimuli, which consist of twochessboardimagespresented on a LCD screen and flashing at particular frequencies, respectively, and the P300 stimuli, which include nine characters and locate between the SSVEP stimuli. We extract the fundamental frequencies and second harmonics of the frequencies of the SSVEP stimulifrom EEG data as features, andapply a Fisher's linear discriminant analysis to build the classifier. The experimental results from three subjects show that the SSVEP potentials can be elicited well under the hybrid stimuli of SSVEP and P300, and the P300 stimuli under the hybrid condition have no significant effect on the SSVEP BCI. This work laysa foundationfor developing a hybrid BCIbased on SSVEP and P300.

In this paper, to develop a mobile robot by using a BCI to issue a motion command including turning left and right and going forward, we propose a new P300 Brain-computer interface (BCI). Three kinds of P300 visual stimuli were first designed. And then we developed the corresponding BCI systems based on the three stimuli and compared their performance by using a linear discriminant analysis classifier with the features selected from EEG potentials via principal component analysis. Experimental results from three participants suggest that the BCIs with the last two kinds of stimuli can select a desired command with over 80% accuracy in about four seconds of selection time, showing that using the proposed BCI to control a robot is feasible.

This paper addresses the design and the experiment of a new E-field probe with small size and ultra wide band. The probe uses an electrical small monopole antenna to sense the outer E-field. A preamplifier based on a HEMT (High Electron Mobility Transistor) increases the probe's sensitivity and the input impedance. A DFB-LD (Distributed feedback laser) modulated by the field strength is employed to transmit the electrical signal to optical signal. The probe was tested in a TEM cell and proved to be able to use in High Power Microwave measurement.

Phase sequence and phase checking is the continuous job in the Distribution system. In this paper we discuss the method of design of virtual phase sequence and phase check meter. The Prototype works steadily and dependably.

Filtering of laser scanner data is important for effective and fast environment recognition of autonomous vehicle in unstructured environment where many uncertainties caused by complex terrain and various obstacles lead to the invalid data existing in laser scanner data. Traditional filtering methods are easy to cause false noise detecting in the complex and changing environment, because not only the size of the filter window is predetermined but also noise is determined under only one condition. A new filtering method is proposed in this paper. In this method, the size of filter window is adaptively determined by boundary points marked in the process of removing the isolated points which divide apparent different points into different groups carrying on a simple pre-classification. Meanwhile, twice threshold comparison is used to judge whether the point is noise or not. Experimental results comparing this method with traditional methods in unstructured environment show that our method reduces the probability of false noise detecting, which brings about the protection of edge and detail information.

CMOS image sensors transmit images by photoelectric conversion and consist of several key analog circuits. We discuss the structure and the overall operating principles of typical CMOS image sensor, and conduct analysis of the key analog circuits function. We carry out the discussion of the theoretical and mathematical principles of those circuits and then we design certain improved key analog circuits on the base of the SMIC 0.5µm 2P3M CMOS craft, such as the pixel cell array circuit, the correlated sampling circuit as well as the Bandgap reference circuit. Our improved designs have improved characteristics such as high output swings, high temperature stability and relatively low room occupation.

Clustering of laser scanner data is a key part of obstacle recognition of automated vehicle with laser scanner by which efficient obstacle determination and fast environment understanding will be achieved through the analysis of the several classes not amounts of data. Traditional clustering methods of laser scanner data based on hard-threshold principle can not meet the requirements of unstructured environment where the topography is complicated and changeable; unknown obstacles are not only complex but also various in kinds. A new adaptive clustering method of laser scanner data is presented in this paper. Little probability event principle is introduced to the nearest adjacent point clustering where every threshold is not fixed and set in advance but is obtained adaptively and effectively by little probability event principle which can reflect the overall change of a class. The new adaptive clustering method is applied for the clustering of ibeo LUX2010 laser scanner data for automated vehicle obstacle recognition in unstructured environment considering both the experimental conditions and the own characteristics of laser scanner. Experimental results show that, compared with the traditional nearest adjacent point clustering based on hard threshold principle, the new adaptive method can characterize the obstacle and the environment information by classes more efficiently and better in unstructured environment meanwhile keep fast enough to meet the real time request of the recognition system.

On the basis of the theory of a generalized system, an optimal investment control strategy model for government has been developed. In order to provide a scientific basis for government to formulate a macroeconomic control policy, the model investigates the balance of total supply and aggregate demand through an adjustment in investment decisions realizes a sustainable and stable growth of the national economy. The optimal investment decision function proposed by this study has a unique and specific expression, high regulating precision and computable characteristics.

Various image filtering strategies have been effective, but most algorithms are to fail in general and create artifacts or remove image fine structures. Hence we propose a novel image filter based on grey polynomial model(GPM) to reduce image noise. The basic conceptions of grey polynomial model are described. The new image filter adopts each noisy pixel's neighborhoods to rectify its intensity value. In addition, the QR decomposition is used to enhance the algorithm's robustness. The experimental results demonstrate that the proposed algorithm, can do better than some more popular denoising methods, such as the median filter and GM(1,1) model, on the removal of low-density impulse noise. Finally, the improved algorithm can effectively eliminate image noise and achieve better denoising effect from both subjective and objective aspects, while preserve the image's textures and edges.

This paper presents a study of satellite-based aquatic petroleum pollution monitoring mode. First, four field data collections were made in 2006 and 2008. The sampling sites were in where the rivers were all threaten by petroleum pollution. The most important feature of water spectra in visible and near-infrared wavelength influenced by petroleum pollution was discovered after analysis in detail. It exhibited three peaks and two troughs in the water spectral curves influenced by petroleum contamination. Secondly, the spectral data collected from the polluted rivers were processed to remote reflectance as corresponding bands of ENVISAT/MERIS. At the same time the detecting wave of MERIS data obtained above were taken into account. And then the MERIS-based monitoring mode of petroleum pollution concentration was established according to Fisher multi-class discriminant rule and combining with corresponding water quality data. Finally, the mode was applied to the ENVISAT/MERIS data overpass on June 5, 2007 in order to obtain the distributing image of petroleum pollution concentration in the rivers. By mean of the image the temporal change and spatial distribute of the petroleum concentration within the rivers were analyzed.