互信息域中的无参考图像质量评价

目的无参考图像质量评价是近几年来的研究热点,具有深远的现实意义和广泛的应用价值,提出一种基于互信息的无参考图像质量评价方法。方法该方法使用原始自然图像及其对应的规范化亮度图像和局部标准差图像作为输入,利用自相关互信息对输入图像邻近像素间的相关性进行量化,并引入多尺度分析得到图像在两个尺度上的互信息特征,最后使用支持向量机(SVM)在LIVE图像数据库上训练学习,从而对多类失真图像进行客观质量评价。结果在LIVE图像数据库中对本文算法进行性能测试,实验结果显示该算法得到的评价结果与人眼主观评价结果之间的平均相关系数高达0.93,总体分类准确率达到79%,性能足以与当前主流的全参考、无参考方法相竞争。结论本文方法有别于传统的基于变换的无参考图像质量评价方法,将着眼点放于自然图像邻近像素之间的固有联系上,并取得了较好的实验效果。由于没有使用图像变换并从全局域进行考虑,本文方法具有较低的时间复杂度。     

Performance analysis for range expansion in heterogeneous networks

In this paper, the uplink and downlink performance of a two-tier heterogeneous network （HetNet） with range expansion （RE） is analyzed. Different from previous works, base stations （BSs） are modeled as a spatial Poisson point process. This model can not only capture the irregularity of the BS deployment but also lead to closed form expressions. With the analytical results, the impacts of several key parameters on the coverage and throughput performance are investigated. Our results justifies the effectiveness of RE and Pico deployment. Based on our analyses, certain attention should be made on the RE design to balance the uplink and downlink performance. Our analyses are corroborated by simulations and can thus serve as a flexible and quick guidance for system design without complicated and time-consuming system-level platform construction and simulation.     

Crop residue,manure and fertilizer in dryland maize under reduced tillage in northern China: I grain yields and nutrient use efficiencies

The rapidly increasing population and associated quest for food and feed in China has led to increased soil cultivation and nitrogen (N) fertilizer use, and as a consequence to increased wind erosion and unbalanced crop nutrition. In the study presented here, we explored the long-term effects of various combinations of maize stover, cattle manure and nitrogen (N) and phosphorus (P) fertilizer applications on maize (Zea mays L.) yield and nutrient and water use efficiencies under reduced tillage practices. In a companion paper, we present the effects on nutrient balances and soil fertility characteristics. The ongoing factorial field trial was conducted at Shouyang Dryland Farming Experimental Station in northern China from 1993 onwards. The incomplete, determinant-optimal design comprised 12 treatments, including a control treatment, in duplicate. Grain yields and N, P, and potassium (K) uptakes and N, P and K use efficiencies were greatly influenced by the amount of rain during the growing season (GSR), and by soil water at sowing (SWS). There were highly significant interactions between GSR and added stover and manure, expressed in complex annual variations in grain yield and N, P and K use efficiencies. Annual mean grain yields ranged from 3,000 kg ha⁻¹ to 10,000 kg ha⁻¹ and treatment mean yields from 4,500 kg ha⁻¹ to 7,000 kg ha⁻¹. Balanced combination of stover (3,000–6,000 kg), manure (1,500–6,000 kg) and N fertilizer (105 kg) gave the highest yield. Stover and manure were important for supplying K, but the effects differed greatly between years. Overall mean N recovery efficiency (NRE) ranged from 28% to 54%, depending on N source. NRE in wet years ranged from 50% to 90%. In conclusion, balanced combinations of stover, manure and NP fertilizer gave the highest yield and NRE. Reduced tillage with adding stover and manure in autumn prior to ploughing is effective in minimizing labor requirement and wind erosion. The potentials of split applications of N fertilizer, targeted to the need of the growing crop (response farming), should be explored to further increase the N use efficiency.     

Modeling impact of culture facilities on hydrodynamics and solute transport in marine aquaculture waters of North Yellow Sea

An increasing number of marine aquaculture facilities have been placed in shallow bays and open sea, which might significantly affect hydrodynamic and solute transport processes in marine aquaculture waters. In this study, a coupled hydrodynamic and solute transport model was developed with high-resolution schemes in marine aquaculture waters based on depth-averaged shallow water equations. A new expression of drag force was incorporated into the momentum equations to express the resistance of suspended culture cages. The coupled model was used to simulate the effect of suspended structures on tidal currents and the movement of a contaminant cloud in the marine aquaculture of the North Yellow Sea, China. The simulation results showed a low-velocity area appearing inside the aquaculture cage area, with a maximum reduction rate of velocity close to 45% under high-density culture. The results also showed that tidal currents were sensitive to the density of suspended cages, the length of cages, and the drag coefficients of cages. The transport processes of pollutants inside aquaculture facilities were inhibited away from the vicinity of the culture cage area because of the diminished tidal currents. Therefore, the suspended cages significantly affected the transport processes of pollutants in the coastal aquaculture waters. Furthermore, the reduced horizontal velocity significantly decreased the food supply for the aquaculture areas from the surrounding sea.     

面向智能驾驶的平行视觉感知：基本概念、框架与应用

目的 视觉感知技术是智能车系统中的一项关键技术,但是在复杂挑战下如何有效提高视觉性能已经成为智能驾驶领域的重要研究内容。本文将人工社会（artificial societies）、计算实验（computational experiments）和平行执行（parallel execution）构成的ACP方法引入智能驾驶的视觉感知领域,提出了面向智能驾驶的平行视觉感知,解决了视觉模型合理训练和评估问题,有助于智能驾驶进一步走向实际应用。方法 平行视觉感知通过人工子系统组合来模拟实际驾驶场景,构建人工驾驶场景使之成为智能车视觉感知的“计算实验室”;借助计算实验两种操作模式完成视觉模型训练与评估;最后采用平行执行动态优化视觉模型,保障智能驾驶对复杂挑战的感知与理解长期有效。结果 实验表明,目标检测的训练阶段虚实混合数据最高精度可达60.9%,比单纯用KPC（包括：KITTI（Karlsruhe Institute of Technology and Toyota Technological Institute）,PASCAL VOC（pattern analysis,statistical modelling and computational learning visual object classes）和MS COCO（Microsoft common objects in context））数据和虚拟数据分别高出17.9%和5.3%;在评估阶段相较于基准数据,常规任务（-30°且垂直移动）平均精度下降11.3%,环境任务（雾天）平均精度下降21.0%,困难任务（所有挑战）平均精度下降33.7%。结论 本文为智能驾驶设计和实施了在实际驾驶场景难以甚至无法进行的视觉计算实验,对复杂视觉挑战进行分析和评估,具备加强智能车在行驶过程中感知和理解周围场景的意义。     

基于RISC-V处理器的卷积加速SoC系统设计

为提高卷积神经网络（CNN）的计算效率和能效,以8 bit定点数据作为输入,设计一个支持激活、批标准化以及池化等CNN网络中常见计算类型的卷积加速器,优化循环计算顺序并将其与数据复用技术相结合,以提高卷积计算的效率。基于软硬件协同设计思想,构建包含RISC-V处理器和卷积加速器的SoC系统,RISC-V处理器基于开源的指令集标准,可以根据具体的设计需求扩展指令功能。将该SoC系统部署在Xilinx ZCU102开发板上,RISC-V处理器和卷积加速器分别工作在100 MHz和300 MHz频率下,测试结果表明,该加速器的算力达到153.6 GOP/s,运行VGG16网络进行图片推理计算时加速效果较好。     

NVRC:一种面向NVM的写限制日志方案

非易失性内存(Non-Volatile Memory,NVM)具有支持按字节寻址、持久性、存储密度高、读写延迟低等特点,因此成为解决DRAM(Dynamic Random Access Memory)容量有限问题的首选技术。随着数据库系统中NVM的引入,传统的日志技术需要考虑如何适应NVM特性。首先总结了已有的面向NVM的日志技术研究,进而提出了一种尽可能限制NVM写操作的数据库日志方案NVRC(Non-Volatile Record-updating with Cacheline)。文中提出了结合异地更新和原地更新的日志管理方案。具体而言,NVRC在异地更新的“影子记录”的基础上,引入了“缓存行原地更新”策略,并通过代价分析选择合理的日志更新策略,从而减少对NVM的写操作。采用DRAM模拟NVM的方式在YCSB测试负载上进行了实验,并对比了NVRC与传统的WAL(Write Ahead Log)以及NVM感知的PCMLx(PCMLoggingx)方法。结果表明,NVRC的NVM写次数在修改均匀的情况下比WAL和PCMLx分别减少了54%和17%,同时更新性能分别提升了59%和10%。     

Autoencoder-based anomaly detection for surface defect inspection

In this paper, the unsupervised autoencoder learning for automated defect detection in manufacturing is evaluated, where only the defect-free samples are required for the model training. The loss function of a Convolutional Autoencoder (CAE) model only aims at minimizing the reconstruction errors, and makes the representative features widely spread. The proposed CAE in this study incorporates a regularization that improves the feature distribution of defect-free samples within a tight range. It makes the representative feature vectors of all training samples as close as possible to the mean feature vector so that a defect sample in the evaluation stage can generate a distinct distance from the trained center of defect-free samples. The proposed CAE model with regularizations has been tested on a variety of material surfaces, including textural and patterned surfaces in images. The experimental results reveal that the proposed CAE with regularizations significantly outperforms the conventional CAE for defect detection applications in the industry.     

基于深度学习的图像本征属性预测方法综述

真实世界的外观主要取决于场景内对象的几何形状、表面材质及光照的方向和强度等图像的本 征属性。通过二维图像预测本征属性是计算机视觉和图形学中的经典问题,对于图像三维重建、增强现实等应 用具有重要意义。然而二维图像的本征属性预测是一个高维的、不适定的逆向问题,通过传统算法无法得到理 想结果。针对近年来随着深度学习在二维图像处理各个方面的应用,出现的大量利用深度学习对图像本征属性 进行预测的研究成果,首先介绍了基于深度学习的图像本征属性预测算法框架,分析了以获得场景反射率和阴 影图为主的本征图像预测、以获得图像中材质 BRDF 参数为主的本征属性预测及以获得图像光照相关信息为主 的本征属性预测 3 个方向的国内外研究进展并总结了各自方法的优缺点,最后指出了图像本征属性预测的研究 趋势和重点。     

High-Precision Wideband Phase-Derived Velocity Measurement for Micro-Motion Extraction

A phase-derived velocity measurement method is proposed in a wideband coherent system, based on a precise echo model considering the inner pulse Doppler effect caused by fast moving targets. The Cramer-Rao low band of velocity measurement precision is deduced, demonstrating the high precision of the proposed method. Simulations and out-field experiments further validate the effectiveness of the proposed method in high-precision measurement and micro-motion extraction for targets with weak reflection intensity. Compared with the long-time integration approaches for velocity measurement, the phase-derived method is easy to implement and meets the requirement for high data rate, which makes it suitable for micro-motion feature extraction in wideband systems.