Search

With the development of artificial intelligence research, the convolutional neural networks (CNN) triumphantly continue to conquer diverse visual tasks, with lots of technical and practical problems emerging. When the detection networks need to meet the high-speed processing requirement and the low power consumption constraint in edge computing, dedicated circuits and algorithm optimization are necessary. In this brief, we propose a non-maximum suppression (NMS) coprocessor that cooperated with a Vision Processor (VP) to speed up the detection. It contains a parallel process unit and dynamic memory management to accelerate NMS processing. Meanwhile, the software workflow provides a comprehensive method to convert the one-stage detection network to our processor in a post-training manner. The experimental results show that our co-design solution reaches 166 FPS to run a MobileNet-SSD, promising to deploy in edge-computing scenarios. Moreover, the detection accuracy on the Airbus dataset is 88.6%, favorable in practical applications. [ABSTRACT FROM AUTHOR]

A tunable pulsed dual-wavelength laser is presented based on an Nd:YVO4/Nd:GdVO4 combined crystals pair as amplifying medium. By employing pulse width adjustment of an 808 nm laser diode pumping, and a Cr4+:YAG crystal for Q-switching, synchronized dual-wavelength laser pulses were obtained at 1063 nm and 1064 nm. Time jitter between the two wavelength pulses is reduced to about 10% of the laser pulse duration. Adjusting the pump beam operation parameters and tuning the crystals pair steady-state temperature between 10 °C and 50 °C, the frequency separation between the dual-wavelength laser pulses gradually decreases from 344.11 GHz to 305.99 GHz. The relative peak powers and the pulses time synchronization remain stable during their frequency separation tuning. [ABSTRACT FROM AUTHOR]

Due to the shortcomings of the weakly supervised and fully supervised object detection (i.e., unsatisfactory performance and expensive annotations, respectively), leveraging partially labeled images in a cost-effective way to train an object detector has attracted much attention. In this paper, we formulate this challenging task as a missing bounding-boxes’ object detection problem. Specifically, we develop a pseudo ground truth mining procedure to automatically find the missing bounding boxes for the unlabeled instances, called pseudo ground truths here, in the training data, and then combine the mined pseudo ground truths and the labeled annotations to train a fully supervised object detector. Furthermore, we propose an incremental learning framework to gradually incorporate the results of the trained fully supervised detector to improve the performance of the missing bounding-boxes’ object detection. More importantly, we find an effective way to label the massive images with limited labors and funds, which is crucial when building a large-scale weakly/webly labeled dataset for object detection. The extensive experiments on the PASCAL VOC and COCO benchmarks demonstrate that our proposed method can narrow the gap between the fully supervised and weakly supervised object detectors, and outperform the previous state-of-the-art weakly supervised detectors by a large margin (more than 3% mAP absolutely) when the missing rate equals 0.9. Moreover, our proposed method with 30% missing bounding-box annotations can achieve comparable performance to some fully supervised detectors. [ABSTRACT FROM AUTHOR]

Energy consumption and delay in end-to-end transmission, two significant cost metrics for guiding the design of routing protocols, are often relevant and even conflicting with each other. Therefore, it naturally becomes important to identify the tradeoff between them. In this paper, we investigate the joint design of routing and power control over unreliable communication links in multi-hop wireless networks with energy-delay tradeoff. Two popular retransmission schemes, hop-by-hop (HBH) scheme and end-to-end (E2E) scheme, are adopted to achieve the reliable transmissions over unreliable links. We model the process of packet transmission in HBH or E2E scheme as a random walk and then calculate the expected energy consumption and the expected delay needed to forward a packet. An expected cost function is defined to capture the tradeoff between energy consumption and delay. Due to the correlation between cost function and transmit power assigned on each link, power control technique is necessary to be integrated into the minimum expected cost routing (MECR) for each retransmission scheme. We prove that the optimal power assignment for each link in a candidate path, which minimizes the expected cost for packet transmission, uniquely exists. Then, the MECR algorithms are proposed for both HBH scheme and E2E scheme to find the optimal routing paths. Simulations illustrate that our proposed protocols not only attain the tradeoff between energy consumption and delay, but also outperform the existing protocols in terms of energy-efficiency and delay. [ABSTRACT FROM PUBLISHER]

In a large and complex distributed sensor network, the individual sensor directly communicates only with its neighboring sensors. Despite being restricted to such local communications, the network itself should be self-organizing to maintain certain desirable properties, such as network connectivity and lifetime maximization. Topology control is a technique to determine the optimal transmit parameters for each sensor, such that the network topology has the best possible network performance. In this paper, we investigate a dynamic topology control scheme to improve the network lifetime for wireless sensor networks in the presence of selfish sensors. A non-cooperative game aided topology control approach is developed for designing energy-efficient and energy balanced network topologies dynamically. Each sensor in the topology control game tries to minimize its unwillingness for constructing a connected network according to its residual energy and transmit power. We prove the existence of Nash equilibrium (NE) and demonstrate that the NE is Pareto optimal as well. Specifically, a fully distributed algorithm—topology control with lifetime extension (TCLE)—is proposed to construct dynamic network topologies. The topologies derived by TCLE algorithm can improve the network lifetime significantly as compared with existing algorithms. Simulations demonstrate the efficiency of our TCLE algorithm. [ABSTRACT FROM PUBLISHER]