Search

In order to solve the problem of particle dilution in traditional particle filter (PF) estimation of Ni-Cd Alkaline battery state of charge (SOC), a bat algorithm optimized particle filter was proposed to estimate SOC. The particles are represented as bat individuals, and the predatory process of bat population is simulated to solve the problem of particle dilution in particle filter technology; the theoretical model of battery state space is constructed by combining the second-order Thevenin battery model, and the related parameters of battery are identified; the SOC is estimated by using BA-PF algorithm and standard PF algorithm under pulse current condition. The experimental results show that, compared with the traditional PF algorithm, the SOC estimation accuracy of Ni-Cd alkaline battery based on BA-PF is less than 2%, and it has good adaptability and stability for nonlinear and non-Gaussian characteristics of Ni-Cd alkaline battery.

Current mainstream multi-object tracking (MOT) algorithms aim to maintain the identities of targets by data-association. However, the accuracy of multi-object tracking would be affected by the unreliability results of detectors. Simultaneously, the separation of motion module and appearance module cannot fully exploit affinity features that will increase computational complexity. To address these issues, we propose a multi-task tracking framework including the United Siamese Network (USN) and Candidate-Refreshing (CR) model. The USN integrates motion affinity and appearance affinity into an end-to-end network. Such design can pay attention to joint learning of similarities between motion patterns and appearance features and enhance the ability to distinguish similar targets in complex environments. The CR model aims to combine detection candidates with tracking candidates, and compensate for unreliable detection results by scoring and regressing the bounding boxes. In addition, we equip our framework with a tracklet confidence function, which is based on the spatialtemporal information. This function can determine the vitality of unmatched tracklets and further alleviate the influence of detectors. Experiments on the MOT16 and MOT17 benchmark datasets show that our framework has achieved outstanding performance on online MOT algorithms.

Learning a powerful distance metric is the key component of image retrieval. Recently, deep metric learning has been an active research topic for image retrieval. However, most existing metric learning approaches treat all the input images equally and learn all image embeddings at equal lengths. These methods ignore those easy examples that can be encoded as the shorter features, which is search-inefficient. We propose a simple but efficient variable-length metric learning method for search efficiency, in which the different query samples have different feature-length. First, we propose to learn the ranked (prioritized) list of features. The more distinguishing features, the higher the rank. We show that the proposed prioritized features can be used to perform fast retrieval in different feature-length configurations. Further, we propose an adaptive feature-length selection policy to determine the amount of feature-length for each query sample. Extensive experiments are conducted on three benchmark datasets. The results demonstrate that the proposed method can reduce computational costs without incurring a decrease in accuracy.

Tracking multiple objects is a challenging task in time-critical video analysis systems. In the popular tracking-by-detection framework, the core problems of a tracker are the quality of the employed input detections and the effectiveness of the data association. Towards this end, we propose a multiple object tracking method which employs a single object tracker to improve the results of unreliable detection and data association simultaneously. Besides, we utilize maximum weight clique graph algorithm to handle the optimal assignment in an online mode. In our method, a robust single object tracker is used to connect previous tracked objects to tackle the current noise detection and improve the data association as a motion cue. Furthermore, we use person re-identification network to learn the historical appearances of the tracklets in order to promote the tracker’s identification ability. We conduct extensive experiments on the MOT benchmark to demonstrate the effectiveness of our tracker.

In this paper, we propose a novel polarization beam splitter (PBS) based on a 3-waveguide directional coupler (DC) with a Si3N4-loaded horizonal slot waveguide as the bridge. The performance of the device can be further improved by adding a filter at the thru port.

With the explosive growth of population and the development of technology, face recognition has been widely applied to various fields. At the same time, there are problems of low recognition speed and accuracy. Therefore, this paper proposes a real-time face recognition system based on edge computing, and implements this system on the development board. The system has excellent performance, high recognition accuracy and low price, which fully exploits the advantages and characteristics of edge computing. This is a good solution for real-time face recognition systems.

Nowadays, teenagers’ vision and cervical health are largely affected by their sitting posture but most of them don’t get into good sitting habits. With the development of wearable smart devices, we have come up with a more effective solution to this problem. A posture reminding system designed with two kinds of sensors and processor Arduino is introduced in this paper. We detect attitude angle and distance from torso to desk using two sensors and send the data to a microprocessor. Then by using Kalman filtering algorithm, we realize the data fusion analysis. To make our device smarter, we apply Genetic Algorithm-Back Propagation neural network for posture classification which shows a higher accuracy compared to traditional classification algorithms. Trained by hundreds of sets of data, the accuracy of our system has reached 99.58% which shows this classification scheme is very effective.

The initial cluster centers of traditional K-means algorithm are randomly selected from spatially distributed data samples. This procedure may significantly affect the final clustering outputs and is unable to ensure a high-quality solution. This paper attempts to improve the quality of solution and the efficiency of clustering through selecting initial cluster centers based on hotspot analysis. An algorithm is developed to identify $K$ hotspots as initial cluster centers. The proposed algorithm is compared to three existing methods in our experiments. Our results demonstrate that our method can generate similar but more stable clustering results with less number of iterations than others.

Where to establish new facilities has received an increasing focus among researchers. Approaches to this question depend on at least a set of three elements, namely representation of demand units, a distance metric, and a location-allocation model. Yet, the choice of these parameters may potentially generate different results leading to significant location errors. The aim of this paper is to quantify the impacts of demand units and distance metric on final location-allocation layout by applying the L-median model to locate one new facility in Spring Hill, Florida, and then provide reference when making the allocation decisions through presenting a picture to illustrate how significant the error is when different decisions are made. The results indicate that distance metric is the most significant factor. Among the four examined metrics, the closest-point distance performs the best than other metrics. On the other hand, the representation of demand units plays a less important role in our case. This research provides guidance on appropriate selection of parameters in location-allocation analysis. One conclusion is that fine demand units are adapted to accurate distance metrics, while coarse demand units to less accurate distance metrics.

Deep Learning is an emerging field in Artificial Intelligence that uses biologically inspired neural networks to recognize patterns in the natural world. These neural networks have an amazing ability to process large amounts of data and learn from them. Recurrent Neural Networks (RNN) are used in applications involving natural language processing like text translations and text generation. This research evaluates the effectiveness of a RNN to be able to automatically generate programming code. Programming languages are different from natural languages in that they have unique structure and syntax. The goal for this research is to conduct experiments on a character RNN model with for three programming languages; Java, Python and C#, and evaluate the results by testing and analyzing the ability for the RNN to automatically produce code that is able to compile.

Geographic Information System (GIS) has a strong ability to visualize and spatial analysis, which monitoring, management, environmental simulation of pollution sources needs. In this paper, in support of GIS we design the goal, framework, structure of Shanghai Key Pollution Sources Management Information System (SKPSMIS) and implemented it. During developing SKPSMIS, enterprise GIS products, ArcGIS Server and Skyline, are used. SKPSMIS has the functions of real-time monitoring and alarm, query statistics, GIS analysis and 3DGIS display, which provides a powerful tool of key sources of pollution monitoring and management.

The role of laser scanning technology in data collection and virtual environment modeling has been long recognized, especially for vehicle-borne laser scanning system (VBLS) which consists of a vehicle equipped with laser range scanners, CCD cameras and positioning devices. Up to now, most researches on vehicle-borne laser data are focused on the extractions of buildings, trees, etc., in correspondence with that on airborne LiDAR data. In this paper, instead, extraction of street lamps becomes our objective and a novel approach is proposed to fulfill it. An experiment is conducted to validate the proposed approach. Comparing with the images collected by the VBLS and real scenario observed from field work, the result indicates that the proposed approach is valid for extracting street lamps in terms of the accuracy of positioning and modeling ground targets.

Improvements in the dielectric properties of composite dielectrics have been previously documented when the filler material used is reduced to nanometric dimensions. While the reasons for this have been traced to the physics and chemistry taking place at the interface, and dramatic changes in the magnitude and dynamics of the internal charge are also known to occur, a clear picture of the exact mechanisms taking place has not emerged. This contribution seeks to compare the direct voltage breakdown of composites formed from biphenyl epoxy resin and titanium dioxide in both nanometric and conventional micron-scale forms with that obtained under impulse voltage conditions. The same materials are subjected to an internal charge analysis using the pulsed electroacoustic technique which shows that, in the case of the nanomaterials, a marked homocharge is formed in front of the cathode which would suggest that the dramatic improvements in voltage endurance seen for these materials may be due to the shielding effect of this negative charge. The finding also suggests that the negative charge is formed as the result of scattering occurring in the nanodielectric which is not present to the same extent in the conventional counterpart.