Search

The spontaneous combustion of coal pile is very easy to cause fires, especially in the closed coal yards. Early detection and disposal of spontaneous-combustion-point (SCP) could avoid large losses. However, the traditional SCP detection methods are confronted with problem of hysteresis, and cannot effectively detect the small SCP in early stage. This paper proposes a novel detection method based on Yolov5s, by using “FPN+ PAN” in the Neck layer to enhance the SCP expressions. To improve the performance of the model in learning the characteristic of small SCP, the block labeling approach is applied in the labeling process. In addition, TensorRT is adopted to accelerate the forward inference of the model, and the deployment test is carried out on Jetson nano. Through comparative analysis, the proposed method outperforms yolov3s and yolov4s in detecting the SCP in very early stage (smoke or flame), and achieves 7 FPS speed in real-time video stream.

The pilot's error is one of the important factors affecting flight safety, which may cause serious consequences. This article mainly discusses the continuous operating deviation and sudden large-scale misoperation during the flight process. For the two types of error models, the controllers are designed to reduce the impact after the misoperation occurs, and simulation verification is performed. Simulation results show that the L1 adaptive controller can effectively suppress the impact of continuous operating deviation on control performance and improve flight comfort; the incremental nonlinear dynamic inverse controller can effectively reduce the impact of large misoperation and restore the original flight state.

The Vision Meets Drone (VisDrone2019) Single Object Tracking challenge is the second annual research activity focusing on evaluating single-object tracking algorithms on drones, held in conjunction with the International Conference on Computer Vision (ICCV 2019). The VisDrone-SOT2019 Challenge goes beyond its VisDrone-SOT2018 predecessor by introducing 25 more challenging sequences for long-term tracking. We evaluate and discuss the results of 22 participating algorithms and 19 state-of-the-art trackers on the collected dataset. The challenge results are publicly available at the website: http://www.aiskyeye.com/. WeexpecttheVisDrone-SOTchallenge to boost the research in single object tracking field.

The Vision Meets Drone Multiple Object Tracking (MOT) Challenge 2019 is the second annual activity focusing on evaluating multi-object tracking algorithms on drones, held in conjunction with the 17-th International Conference on Computer Vision (ICCV 2019). Results of 12 submitted MOT algorithms on the collected drone-based dataset are presented. Meanwhile, we also report the results of 6 state-of-the-art MOT algorithms, and provide a comprehensive analysis and discussion of the results. The results of all submissions are publicly available at the website: http://www.aiskyeye.com/. The challenge results show that MOT on drones is far from being solved. We believe the challenge can largely boost the research and development in MOT on drone platforms.

In this paper, a real-time unmanned aerial vehicle (UAV) tracking method is proposed. This approach builds a target representation from histograms of oriented gradient (HOG) and ColorNames features. Correlation filters have been utilized in tracking recently because of their high efficiency. To better fuse the tracking results from different features, peak-to-sidelobe ratio (PSR) is employed to evaluate robustness of our trackers. A stability measure is proposed, based on the PSR values computed over a short period of time which is also used to predict object position. Additionally, we show that the proposed PSR stability enables our tracking method to be robust to various appearance variations. The method is carried out on five UAV tracking datasets and achieves appealing results comparable to state-of-the-art trackers but at a lower computational complexity.

The emergence of a large number of video data puts forward higher requirements on traditional video transmission technology. The new streaming media technology based on HTTP dynamic adaptive streaming DASH transmission protocol has become an important research direction of video services. How to overcome the unstable characteristics of wireless links in a limited bandwidth, achieve high-quality intelligent transmission of video, and obtain optimal user quality of experience (QoE), has become an urgent problem to be solved. This paper abandons the traditional streaming media adaptive transmission method, and combines neural network and particle swarm optimization algorithm to design a new intelligent transmission scheme. The particle swarm optimization algorithm obtains the optimal transmission parameters of QoE, and the model established by neural network predicts the optimal one. The system sets parameters to ensure video service quality under limited bandwidth and large network fluctuations in the wireless network.

At present, many deep neural networks are applied to image recognition. But most of them are based on real-valued operations and represents. Since the algorithm of complex operation has been put forward, we apply the VGG model to the complex domain in the paper. We provide the advantages which the complex-valued network possesses in terms of the depth and width of networks by calculating. When possess the same parameters, the complex-valued network is deeper and wider than the real-valued network. We test both complex-valued VGG network and real-valued one on image recognition. Experiments show that the complex-valued VGG network has better performance comparing with the traditional real-valued VGG network in terms of stability and convergence speed.

To reduce the complexity associated with application-specific tuning of wireless sensor networks (WSNs), dynamic profiling enables an accurate view of an application’s runtime behavior, such that the network can be reoptimized at runtime in response to changing application behavior or environmental conditions. However, the dynamic profiling must be able to accurately capture application behavior without incurring significant runtime overheads. Since application- and sensor-specific constraints dictate the profiling requirements and tolerated overheads, designers require design assistance to quickly evaluate and select appropriate profiling methodologies. To increase designer productivity, we formulate profiling methodology design guidelines based on extensive evaluation and analysis of a variety of profiling methodologies suitable for dynamically monitoring WSNs with respect to network traffic overhead, power, and code impacts associated with each method. While energy consumption increases are reasonable, ranging from 0.5% to 2.6%, network traffic, code size, and computation time overheads can be as high as 66.2%, 75.9%, and 136.6%, respectively. Our results show that these overhead variations are highly application specific, and a single profiling method is not suitable for all types of application behavior, thus necessitating, application-specific profiling methodology customization. To facilitate rapid development of these profiling methodologies, we present a profiler-customization methodology consisting of a code generator module, overhead estimation module, and profile data management module. Using our profiling-customization methodology, designers can rapidly evaluate the overhead of different profiling methodologies, and automatically integrate the most appropriate methodology into the application at design time.

With the development of science and technology, robots have gradually come into our life from industry. Tennis ball collection robot is one of the more prominent representatives. How to find the position of tennis ball accurately is the key. In order to improve the speed and accuracy, we use deep learning and divide the image recognition into two steps. Using AlexNet network model to test whether there is a tennis ball in the picture. If a tennis ball exists, we then use Single Shot multibox Detector (SSD) model to locate the tennis ball, otherwise we use AlexNet to continue to check the next image. Upon verification, we can achieve the desired results.

A stigmergy-based aggregation method for swarm robots is introduced in this paper. The method uses finite state machine and coding-based pheromone to coordinate the behavior of robots to accomplish aggregation tasks collaboratively. The robots behaviors are divided into three states by using the finite state machine. A coding-based pheromone generation rule is designed. Part of the robots leave pheromone trajectories with encoded form in the environment, and the other robots plan their motion according to the pheromone trajectories. Eventually all the robots gather to the specific area. RFID tags are arranged in the environment as the carriers of pheromone. Robots exchange information via the tags as they move over the tags and decide their next motion in accordance with the information obtained from the tags. Simulations and experiments are made to verify the method and the performance of real robots.

In order to overcome the limitation of the traditional bridge detection method, under the background of the unmanned aerial vehicle detection bridge, taking into account the signal without GPS, the information acquisition process and the grid method are combined to establish an environmental model, extremely covered the bottom of the bridge. In the path planning, most of the algorithms can only plan the shortest path, and the algorithm is complex, requires a lot of storage space, and it is difficult to carry out the whole area coverage path planning. Now the unit decomposition method and the two-point search strategy are proposed and applied in this context, so the algorithm has the advantages of simple computation and covering intelligence, It has great potential in the coverage path planning. Simulation experiments show that it can achieve the whole area coverage under the conditions of low power consumption.

Electronic nose is a system based on the sense of smell, which is consisted of several metal oxide semiconductor sensors based on the characteristic odor or volatile components in the air. Its structure includes gas acquisition system, sensor array, signal conditioning circuit and data processing. Compared with other systems, it has the advantages of sensitive detection, simple operation, and high cost performance, etc. However, the materials of sensors are non-linear metal-semiconductor, the obtained odor information must also include non-linear characteristic. Traditional linear feature extraction algorithm can't extract the nonlinear characteristics, so this work proposed four manifold learning algorithms for extracting odor information characteristics. In order to verify feature extraction results, electronic nose measurement data based on the living Chinese mitten crab was processed. The prediction model based on support vector machine (SVM) algorithm according to the guide of the above classification was presented. Total volatile basic nitrogen (TVB-N) of the samples was measured and its results acted as reference information. Experimental results illustrated that the system based on electronic nose with manifold learning can assess the freshness of the living crab.

Optimization of sensor networks relies on accurate profiling information collected about the state of individual nodes and the network as a whole. A single fixed profiling methodology may incur significant overheads on the sensor network or produce inaccurate profiling results due to dynamic changes in application behavior at runtime. Alternatively, reconfiguring the profiling methodology at runtime in response to such changes can help maintain the accuracy of profiling results while minimizing the associated overheads. In this paper, we present a runtime adaptive profiling methodology that can adapt to runtime behavior of the network and preserve the accuracy of profiling data. This runtime adaptive profiling strategy further allows application experts to control the profiling accuracy, thereby providing a mechanism to tradeoff accuracy and overhead. Experimental results demonstrate that network, computational time, and power consumption overheads can be reduced by more than 50% compared to using a fixed profiling methodology while only missing 2% of profiled events.

The variation of modality of a specimen plays an important role in a buckling experiment. However, it is not easy to obtain a whole field measurement by most traditional manners in some particular experimental condition. In this paper, we develop a simple method for out-of-plane displacement measurement on a curved surface using one camera. An algorithm based on LDC (Line Detection using Contours) is used to detect corners. The out-of-plane displacement is then determined by a geometrical method on the captured images. And by using multiple scales method, the problem of the variable prospective shrink rate on a curved surface is simplified. Experiments of buckling specimens with curved surface demonstrate that this method is simple, accurate and suitable for measurement of out-of-plane displacement in the limited experimental conditions.

In this paper, we propose a novel Open Loop Power Control (OLPC) scheme for Long Term Evolution (LTE) uplink. The total transmission power of each user is determined by two parts: the power on one RB which is adjusted according to user's pathloss and the number of Resource Blocks (RBs) assigned to it. Conventional OLPC methods only adjust users' power on one RB and assign RBs equally among all users, which results in problems: users in good channel condition suffer from low data rate with limited number of assigned RBs, while users in bad channel condition cause large inter-sector interference because of their large transmission power. In this paper, we jointly consider power control and resource allocation according to users' channel condition. In the proposed method, users in good channel condition can be assigned with more RBs compared to users in bad channel condition. This method makes efficient use of limited radio resource and reduces interference in system. Simulation results show that the proposed method provides at least 18% spectral efficiency gain compared to the conventional method.

Microwave photonic filter requires high-speed Radio frequency (RF)-optical and optical-RF signal conversion. Light sources and photo-detectors (PDs) are indispensible for such conversions. Monolithic integration of functions of photon emission and detection is demonstrated using epitaxial Ge on Si. The fabricated waveguided PD shows a high bandwidth of 10 GHz with the responsivity of 0.8 A/W and dark current of 150 nA at −1V. The photoluminescence (PL) properties of P-implanted Ge are studied, and an improved annealing process is demonstrated to enhance the light emission by 1 order. Electroluminescence (EL) is obtained from the waveguided n+-Ge light-emitting diode (LED), and the emitted power has been studied as a function of injected current. The successful demonstration of both waveguided PD and LED on single Si chip would be attractive for fabricating a compact size microwave photonic filter with low cost and high integrity.

This paper presents a novel modulation and coding scheme (MCS) selection method to optimize the performance of Voice over Internet Protocol (VoIP) semi-persistent scheduling in Long Term Evolution (LTE) downlink. In VoIP traffic, the packet length little changes during a certain period and the cost of resource for the packet is related to this length. We exploit the characteristics and propose to select a more reliable MCS with the same amount of resource to be allocated. This method can reduce the block error ratio (BLER) at the same time satisfy VoIP Quality-of-Service (QoS) requirements. Through system level simulations, the proposed MCS selection criterion is proved to yield nearly 15% capacity gain in LTE VoIP system compared with the conventional method.

We have used the Pockels effect in ZnTe to demonstrate that the frequency and the phase of a 2.7Terahertz Quantum Cascade Laser can be actively stabilized to the n-th harmonic of the 90MHz repetition rate of a commercial, mode-locked Er-doped fiber-laser. The beating between the stabilized QCL frequency and the harmonic of the repetition rate yielded a signal-to-noise ratio of 80dB in 1Hz bandwidth (limited by the resolution of our spectrum analyzer). The technique demonstrated is inherently broadband, in the sense that it allows to directly stabilize any QCL source provided that its frequency is within the spectral bandwidth of the fs-laser (∼5THz).

The Unmanned Aerial Vehicle (UAV) data link is composed of ground control station and airborne data terminal, all of which communicate in UART mode. In order to achieve multi-serials communication between UAV and ground control station, a method of multi-serials to Ethernet Gateway based on field programmable gate array (FPGA) +network interface chip is given. Firstly, this paper introduced the structure and working principle of system. Then the design of system's hardware and software programming were described, as well analyzed configuration of every module and communication of serial ports in detail. The modules of transfer and receiver were completed by state machine. Finally, the timing simulation of two serial ports was shown. Compared with traditional methods, it possesses the characters of high-integration, low-power, and flexibility. It has been proved in the real test that the design can communicate with computer and work well, and achieve the prospective purpose. It is of much great practical significance and operational benefits.

Telecontrol target vehicle is often used for developing, experimenting, inspection and training of air to ground and ground to ground equipments, generally travels along a special straight orbit, in order to shorten accelerating time, increase the flexibility and reliability of governing speed in limited length and speedup in a short time. This paper attempts to show a kind of implementation of stepless shift gears control for the telecontrol target vehicle system. Firstly, structure, function and principle of system are given. Then, hardware implementation of the modules is given in detail, including measure device of velocity and controller of the gun. Lastly, the software flow is introduced about the realization of stepless shift gears by using singlechip. The target practice showed that functions have been finished for automation shift gears in finite distance and speedup in the shortest time, and this method can reduce the system cost, increase the system agility and reliability.

This paper first analyzes the mathematical model of mobile Ad Hoc networks by considering a weighted value of the ratio between sustainable energy and the total consumed energy of nodes in order to find a balance point between them. And it proposes a solution to the problem of energy efficient and load balance. Finally it presents some improvement suggestions to the AODV protocol.

This paper provides an improved search algorithm for optimal route planning by rebuilding the search area in intelligent transport system. The theory foundation is that the classical Dijsktra algorithm has not any directional feature during searching the optimal path, and the bidirectional Dijsktra has its own limit. Based on the analysis of the two algorithms, the new improved algorithm proposed in this paper reduces the search area and the time complexity. Finally, combining with the practical application of route planning algorithm in intelligent transport system, one example is given. Theory analysis and test result show that the improved algorithm costs less search time and visits less nodes of road network. Besides that, some methods are proposed to handle unexpected disturbance, such as accident or traffic limit. Two models for introducing dynamics in the system are introduced as individual edge dynamics and geometrical objects. These methods are useful in practice.

In order to overcome the limitation of single sensor in vehicle integrated navigation system, cascade fusion architecture is proposed to enhance the reliability of location information. Our research is focus on the algorithm in decisionmaking level of the fusion architecture, which is used to fuse the information from Global Positioning System (GPS), Kalman filter and Map Matching (MM) to get the precise location. The proposed algorithm in this paper utilizes Particle Swarm Optimizer (PSO) to substitute the traditional Back-Propagation (BP) algorithm in training parameters of neural net. It has more generalization capability. Besides that, it converges stably and is resistant to local optima compared with traditional BP. Test result shows that the proposed algorithm can improve location accuracy by making full use of all sensors' information, and it is robust and effective.

Two efficient implementation forms of unscented Kalman filter are described in this paper. They respectively utilize Cholesky factors and modified Cholesky factors to update state covariance, which make them have good numerical stability. Through computational complexity analysis, it is shown that the implementation using Cholesky factors is slower than the standard implementation. Also is shown that in some cases the form using modified Cholesky factors is faster than the standard unscented Kalman filter.

The converse effects of spin photocurrent and current induced spin polarization are experimentally demonstrated in a two-dimensional electron gas system with Rashba spin splitting. Their consistency with the strength of the Rashba coupling as measured for the same system from beating of the Shubnikov-de Haas oscillations reveals a unified picture for the spin photocurrent, current-induced spin-polarization, and spin-orbit coupling. In addition, the observed spectral inversion of the spin photocurrent indicates a system with dominating structure inversion asymmetry.