雨天变电站设备智能视觉检测算法与实时实现技术

视频的变电站户外设备监控是保障电力系统安全运维的重要手段,但传统的目视监控效率低、人力成本高,因此无人化智能视觉检测得到广泛关注。变电站户外监控设备在采集图像时,不可避免会遭受雨天等恶劣天气影响,导致设备检测性能严重下降。针对雨天场景下的变电站设备检测给出了软硬件系统解决方案,主要围绕雨天变电站设备视觉检测算法和嵌入式AI实时部署展开研究。首先,提出一种基于细节增强的图像去雨网络(DERN),利用雨条的高频特性作为注意力权重指导去雨,进一步将其与YOLOv4目标检测算法进行联合优化,从而实现对雨天稳健的变电站设备检测算法;基于昇腾310 AI处理器构建嵌入式边缘端智能计算系统,从算子优化、多线程并行等方面提升计算性能。实验表明,在雨天场景下变电站设施设备目标检测精度从实施前的53.4%提高到82.1%,1080P视频运行速率达到30 f/s,验证了所提算法和系统的有效性,有效提升了变电站设备全天候视觉智能监控能力。

一种基于REC.2020高清色域的视频级显微高动态范围成像方法

高动态范围成像已经成为数字显微镜发展的一大趋势。目前视频级高动态范围显微成像技术由于计算复杂度较高,还无法做到实时处理和显示。提出利用高清电视领域中的REC.2020色域变换,结合GPU实时加速处理技术,成功实现了高动态范围显微图像视频流的实时显示,图像具有良好的色彩还原性和细节恢复性。实验结果表明,设计的算法不仅能够提供稳定的高清、高动态范围视频,其细节和色彩还原性也有了很大的提高,突破了传统的高动态范围图像技术在成像速度和质量方面的限制。

基于RAPIDS的无参DBSCAN算法

具有噪声的基于密度的空间聚类(Density‑based spatial clustering of applications with noise,DBSCAN)能够发现不同密度和大小的类簇,对噪声也有很好的鲁棒性,被广泛地应用到数据挖掘的任务中。DBSCAN通常需要调整参数MinPts和Eps以达到更优的聚类效果,但往往在搜索最优参数的过程中会影响DBSCAN的性能。本文从两个方面优化DBSCAN,一方面,提出一种无参的方法优化DBSCAN全局参数选择。无参方法利用自然最近邻获得数据集的自然特征值,并将自然特征值作为参数MinPts值。然后,根据自然特征值计算自然特征集合,利用自然特征集合中的数据分布特性,分别采取统计最小值、平均值和最大值3种方式得到Eps值。另一方面,采用集成数据科学实时加速平台(Real‑time acceleration platform for integrated data science,RAPIDS)的图形处理器(Graphics processing unit,GPU)计算加快DBSCAN算法的收敛速度。实验结果表明,本文提出的方法在优化DBSCAN参数选择的同时,取得了与密度峰值聚类(Density peaks clustering,DPC)相当的聚类结果。

Real-time reliability assessment based on acceleration monitoring for bridge

Bridge health monitoring （BHM） has become increasingly significant in the life-cycle of the structure such as maintenance, repair and rehabilitation. It is necessary to use BHM information efficiently to assess the working conditions of the bridge. The main objective of this study is to develop an effective method and establish a framework for the real-time reliability assessment based on BHM acceleration information. The first-passage probability and its further development have been proposed to as- sess the reliability probability. The first-passage probability shows the probability of that a scalar process exceeds a designated threshold during a given time interval. The advantage of the proposed method is the assessment of the real-time reliability probability based on the monitoring information during an assessment reference period. Furthermore, the velocity data and displacement data are calculated from the acceleration monitoring data using the relationships between their power spectral density （PSD） functions. The real-time reliability assessment of Donghai Bridge, which is the first large scale cross-sea bridge in China, demonstrates that the proposed method is efficient and effective.

Real-time pre-processing system with hardware accelerator for mobile core networks

With the rapidly increasing number of mobile devices being used as essential terminals or platforms for communication, security threats now target the whole telecommunication infrastructure and become increasingly serious. Network probing tools, which are deployed as a bypass device at a mobile core network gateway, can collect and analyze all the traffic for security detection. However, due to the ever-increasing link speed, it is of vital importance to offioad the processing pressure of the detection system. In this paper, we design and evaluate a real-time pre-processing system, which includes a hardware accelerator and a multi-core processor. The implemented prototype can quickly restore each encapsulated packet and effectively distribute traffic to multiple back-end detection systems. We demonstrate the prototype in a well-deployed network environment with large volumes of real data. Experimental results show that our system can achieve at least 18 Gb/s with no packet loss with all kinds of communication protocols.

High speed interpolation for micro-line trajectory and adaptive real-time look-ahead scheme in CNC machining

CNC machining plays an important role in mechanical manufacturing.A key issue is to improve the machining feedrate while keeping the machining precision and satisfying the acceleration constraints of the CNC machine.For the consecutive micro-line segments interpolation,the velocities at the junction of two segments are the bottlenecks for the machining efficiency.This paper proposes a multi-period turning method to improve the feedrate at the junctions using the linear acceleration and deceleration mode,which utilizes the maximal acceleration capabilities of the NC machine while satisfying the machining precision.A new and more efficient look-ahead method and a feedrate override method are also proposed to boast the global machining speed.The proposed algorithm has been implemented on Blue Sky NC System,and experimented in real material manufacturing.Compared with several existing algorithms,the current algorithm can improve the manufacturing time ranging from 50% to 180%,depending on the machining parameters,and also results in better machining quality.In addition,the algorithm also satisfies the need of real-time interpolation.

Recent advances in efficient computation of deep convolutional neural networks

Deep neural networks have evolved remarkably over the past few years and they are currently the fundamental tools of many intelligent systems.At the same time,the computational complexity and resource consumption of these networks continue to increase.This poses a significant challenge to the deployment of such networks,especially in real-time applications or on resource-limited devices.Thus,network acceleration has become a hot topic within the deep learning community.As for hardware implementation of deep neural networks,a batch of accelerators based on a field-programmable gate array(FPGA) or an application-specific integrated circuit(ASIC)have been proposed in recent years.In this paper,we provide a comprehensive survey of recent advances in network acceleration,compression,and accelerator design from both algorithm and hardware points of view.Specifically,we provide a thorough analysis of each of the following topics:network pruning,low-rank approximation,network quantization,teacher–student networks,compact network design,and hardware accelerators.Finally,we introduce and discuss a few possible future directions.