Search

This paper aims to address the problem of oriented object detection under the complex background of remote sensing images. To this end, we propose a one-stage object detection method with feature fusion structure, and modify the loss function to enhance the detection of small objects. More specifically, on the basis of the end-to-end one-stage object detection model RetinaNet, the method of gliding the vertices of the horizontal bounding box is used to describe an oriented object. In order to obtain multi-scale context information, we design a feature fusion module. Besides, we propose a novel area-weighted loss function to pay more attention to small objects. Experimental results conducted on the DOTA dataset demonstrate that the proposed framework outperforms several state-of-the-art baselines.

Electro-analytical (E-chem) biosensors offer unique advantages over widely used optical biosensors and can be considered ideal for low-cost, mass-deployable point-of-care (PoC) diagnostic devices [1]. They possess fully electronic and real-time transduction methods and require little or no external instrumentation. Despite these advantages, two fundamental challenges have hampered broad adoption of E-chem biosensors: 1C design and transducer (surface) material suitability. E-chem sensors require both low-noise and high detection dynamic range (DDR) front-end circuits and must also accommodate electrode-electrolyte interfaces with significant PVT variations and temporal drifts [2]–[5]. The absence of CMOS-compatible bio-electronic interfaces with adequate chemical and thermal stability has been another impediment [6].

Traditional clonal selection algorithm gives us some inspiration for many applications, but in the medical image enhancement application we found some defects of this algorithm. So we improved the clonal selection algorithm in three ways. First, we designed the real coding of the MRI brain image, instead of binary coding. Second, we added the mutation distance to control the mutation progress and avoid only the local optimization. In addition, we adjust the clone selection and the mutation together in the Gauss distribution, the uniform distribution, and the chaotic distribution, rather than in only the Gauss distribution. Then we use the real MRI brain images to test the image enhancement of our improved clonal selection algorithm. The experimental results show that our approach outperform the median filtering (MF) and the adaptive template filtering (ATF) in enhancing the MRI brain images.

In the immune responses against the SARS (Severe Acute Respiratory Syndromes), human immune systems are complex intelligent systems, which show good properties such as the self-organizing and adaptivity. Modeling the immune systems has important significance in both immunology and artificial immune system. In order to improve the visualization and readability of the anti-SARS immune system model, the visual tri-tier computational model of the anti-SARS immune system was simulated with NetLogo, which is a multi-agent-based tool. On the other hand, to fight against the SARS disease, the lowresolution medical CT (Computed Tomography) images should be transformed into the high-resolution ones for better SARS analysis. In order to obtain the high-resolution image from some low-resolution chest CT sequence images of a SARS patient, the low-to-high resolution reconstruction was designed and tested in this paper. First, the low-resolution medical images were preprocessed. Then the pretreated low-resolution medical images were registered with the sub-pixel-level image registration techniques. Finally, the POCS (Projections onto Convex Sets) image reconstruction algorithm was designed and tested. We obtained higher entropy and more detail information of the medical images with our approach than the Marcel method, especially for the rotated medical images in our experiments. Multiple-user browser-based experimental results show that the visual NetLogo-based simulation of the immune system is better to understand than the traditional mathematic equation model of the immune system.

Battery parameter identification is a foundation of dynamic estimations for battery performance states, such as state of charge (SOC) and state of health (SOH), which are key factors in ensuring a battery system's effectiveness, safety and reliability. In this study, a real-time and training- free method is developed to identify such parameters as open-circuit voltage, internal resistances, and polarization capacitance of a first- order resistance-capacitance (RC) circuit model on line. The state-space equations are formulated according to model's dynamic characteristics, and an extended Kalman filter (EKF) algorithm is employed for the parameter identification only using the measured current and voltage of batteries. Federal Urban Driving Schedule (FUDS) cycles are performed in validation experiments on LiFePO4 cells. The results demonstrate the accuracy and regression of the proposed algorithm. Additionally, the algorithm was successfully applied in a battery management system (BMS) with a low computational complexity.

A new battery modelling method is presented based on the simulation error minimization criterion rather than the conventional prediction error criterion. A new integrated optimization method to optimize the model parameters is proposed. This new method is validated on a set of Li ion battery test data, and the results confirm the advantages of the proposed method in terms of the model generalization performance and long-term prediction accuracy.

Compared with the traditional video-based traffic surveillance, the highly-mounted video can surveillance a larger range of traffic scene, and the vehicles in the video are rarely occluded. These characters of highly-mounted video-based traffic surveillance system make it more suitable for acquiring the traffic information of a larger traffic area. This paper present a framework of using highly-mounted video to extract the real-time traffic information. A single Gaussian model is used to estimate and update the background, and then the foreground or moving vehicles are detected; Meanwhile, we transform the image coordinate to the world coordinate with camera calibration to calculate such traffic information as the average velocity of vehicles, and the average length of the vehicle queue on the road; A region-prediction-based technique and feature-based method are combined together to track the moving vehicles, along which the traffic information are calculated.

The increasing interest and utilization of Wireless Sensor Networks has increased the requirements of energy saving for battery powered sensor nodes. Even in modern sensor nodes, communication causes the largest part of energy consumption and therefore ways to reduce the amount of data sending are widely concerned. One solution to reduce data transmission is a model-driven data acquisition technique called Derivative-Based Prediction (DBP). Instead of transmitting every measured sample, a sensor node uses algorithms to compute approximated models to represent the measured data. In this work, we developed an algorithm to monitor temperature samples in different environmental scenarios. We also evaluated the algorithm with regard to its efficiency and classified the recorded temperature patterns to enhance the precision. In our tests, the algorithm successfully suppressed up to 99% of data transmissions while the average error of prediction has been kept below 0.1°C.

Traditional search behavior analysis was too single to meet the individual needs of the user information search. This paper introduces the artificial intelligence agent technology analysis of user search behavior, analyzes 5 kinds of user search behavior by the use of multiple agent Intelligent agent technology advantage, and gather five kinds of user search behavior analysis results, output user information satisfaction, personalized service to meet user. To improve the efficiency of the search engines to provide a new method.

State of health (SOH) estimation for Lithium-ion batteries is a core technology of battery management systems, which is a foundation of ensuring system's safety and reliability. An accurate SOH estimation reflects the present fraction of allowable performance left before the end of life. In this paper, quantization standards of SOH are introduced firstly and some suggestions are proposed to solve the inconformity issue of SOH definitions. Then the battery aging phenomenon is interpreted theoretically, which contains the analyses of the aging mechanism, the changes of parameters during aging processes and the factors of accelerating aging. Next, some main methods of the SOH estimation are summarized and compared with each other. And finally, this paper describes the development trend of SOH estimation algorithm.

The traditional Zernike moment-based method is often used for subpixel-level edge detection, but the algorithm still requires manual tuning on selecting threshold, not only inefficient, but also leads to the wrong detection of the edge and affecting the measurement accuracy. In order to overcome this weakness, this paper combines original algorithm with Otsu adaptive threshold algorithm and a new fast algorithm for edge detection is proposed. This fast algorithm can directly obtain the optimal step gray threshold and detect the edge. The detection time is significantly shortened while ensuring the detection results. This algorithm also calculates the error caused by edge model. Experimental results show that the improved algorithm can detect the image edge more effectively and the subpixel location is more accurate after compensating, the detection results are satisfied.

Appropriate monitoring technology for energy storage system plays an important role in electric vehicles. A distributed data acquisition system was developed. The system which communicates with host unit via CAN-bus is composed of 36 data acquisition units which monitoring cells parameters. The distributed topology could effectively improve the speed of data acquisition. A calibration algorithm for current sensor was developed to increase the estimation precision of SOC. The monitoring system was implemented in a super-capacitor pure electric urban bus. Experimental results demonstrated that the distributed data acquisition system performs well in monitoring the system and cell voltage, current and temperature.