基于改进Faster R-CNN与U-Net算法的桥梁病害识别与量化方法

为实现桥梁病害检测的自动化,对基于图像处理技术的混凝土桥梁表观病害的智能识别和尺寸确定方法展开研究.提出基于改进Faster R-CNN算法的病害识别方法,利用K均值聚类和遗传算法对区域候选网络锚框进行优化设计;以裂缝预测区域为基础,提出ResNet34结合U-Net的裂缝形态提取方法,并结合裂缝形态学研究了裂缝像素宽度和长度的确定方法.结果表明:锚框优化设计可改进Faster R-CNN算法的表观病害识别效果,5类常见病害的预测准确率、召回率、平均精确率分别由68.40%、69.87%、74.64%提升到85.40%、83.59%、83.72%;利用病害预测框,结合改进U-Net算法的裂缝像素尺寸计算,可实现裂缝病害尺寸的自动测量;基于改进Faster R-CNN和改进U-Net的方法可实现混凝土桥梁常见病害的智能识别和尺寸量化,从而提高桥梁病害检测效率并促进桥梁技术状况评定的智能化.

改进Mask R-CNN的无人机影像建筑物提取

从无人机影像中自动提取建筑物对城乡规划和管理至关重要,然而,在复杂背景干扰和建筑物外观变化很大的情况下给实例提取带来挑战。因此,提出一种改进的Mask区域卷积神经网络(R-CNN)方法用于无人机影像的建筑物自动实例提取。改进方法以ResNet-101作为特征提取网络,在特征融合网络方面,通过添加自底向上的路径增强整个特征层次的定位能力,同时在特征融合中加入空洞空间金字塔池化模块(ASPP)来提高多尺度能力与改善模型性能。在自制建筑物数据集上的综合实验结果表明,与原始的Mask R-CNN方法相比,改进方法的mAP值提高了2.6%,能够很好地实现无人机影像建筑物实例提取。

Research on microwave radiation features and lunar regolith parameters inversion of the Rümker region

Chang'e-5(CE-5)mission is expected to land and sample in the Rümker region,north of the Oceanus Procellarum.To select optimal sampling points,the microwave radiation features in this mare unit are analyzed,and the dielectric constant and thickness of lunar regolith are retrieved based on Chang'e lunar microwave sounder(CELMS)data.According to the microwave brightness temperature(TB)maps,TB varying with frequency at noon and midnight are different,in which noon TB increases with frequency,but midnight TB does not follow this rule.Moreover,there are differences among TB distributions at the three higher frequencies,especially in the southern part.In addition,a great TB difference between noon and midnight can be seen in Rümker E that shows higher daytime temperature but lower nighttime temperature.The inversion results reveal that the dielectric constant is lower in the west and north,higher in the middle and east,and the largest value occurs in Rümker E.As for the regolith thickness,most areas are uniform and no more than 4 m,with the thickest and thinnest regolith appearing in the northwest and Rümker E,no less than 6 m and about 2.5 m,respectively.Considering the safety and operability of landing and sampling,the northwest with thick regolith is recommended,while the Rümker E should be eluded.

基于改进Faster R-CNN的钢板表面缺陷检测

针对钢板表面缺陷检测过程中出现的检测精度不高、小缺陷定位不准、误检率和漏检率高的问题,提出一种改进的快速区域卷积神经网络(faster region-based convolutional neural network,简称Faster R-CNN)算法.实验结果表明:相对于CNN(convolutional neural network),YOLOv3(the third version of you only look once),SSD(single shot multibox detector)和YOLOv5(the fifth version of you only look once)算法,该文算法的检测平均精度均值最高、缺陷定位最准确、分类成功率最高.因此,该文算法具有有效性.

基于R语言的GGE双标图法在甜菜品种区域试验中的应用

为了明确糖用甜菜品种在全国不同区域试验中参试品种的丰产性、稳产性、适应性及各试验点的区分力和代表力。2020年在全国不同生态区域的9个试验基地,以16个引种的KWS系列糖用甜菜品种为试材,采用基于R语言的GGE双标图法对糖用甜菜品种的产糖量指标进行分析。结果表明,在2020年品种区域试验中,KWS 0015(G16)丰产性最佳;KWS 7748(G6)、KWS 9921(G15)具有较好的稳产性;KWS 0015(G16)适宜种植的区试点最多,具有较强区域适应性,较其他品种高产稳产,为本试验理想品种。此外,呼和浩特(E9)具有较高区分力和较好代表性,是本研究的理想区试点。GGE双标图法对综合分析糖用甜菜品种基因型、环境与基因型交互效应具有科学合理性。

On the criteria to create a susceptibility map to debris flow at a regional scale using Flow-R

Studies on susceptibility to debris flows at regional scale(100-1000 km^2) are important for the protection and management of mountain areas. To reach this objective, routing models, mainly based on land topography, can be used to predict susceptible areas rapidly while necessitating few input data. In this research, Flow-R model is implemented to create the susceptibility map for the debris flow of the Vizze Valley(BZ, North-Eastern Italy; 134 km^2). The analysis considers the model application at local scale for three sub-catchments and then it explores the model upscaling at the regional scale by verifying two methods to generate the source areas of debris-flow initiation. Using data of an extreme event occurred in the Vizze Valley(4 August 2012) and historical information, the modeling verification highlights that the propagation parameters are relatively simple to set in order to obtain correct runout distances. A double DTM filtering-using a threshold for the upslope contributing area(0.1 km^2) and a threshold for the terrain-slope angle(15°)-provides a satisfactory prediction of source areas and susceptibility map within the geological conditions of the Vizze Valley.

Projection of PM_(2.5) and Ozone Concentration Changes over the Jing-Jin-Ji Region in China

Based on GISS-E2-R model simulations, the changes in PM2.5 and ozone concentrations during 2016– 35 are analyzed over the Jing-Jin-Ji region under different future emissions scenarios: 2.6, 4.5, 6.0, 8.5 Representative Concentration Pathways scenarios(RCP2.6, RCP4.5, RCP6.0, and RCP8.5), compared to the baseline periods of 1851–70(pre-industrial) and 1986–2005(present day). The results show that PM2.5 increases under all emissions scenarios, with the maximum value occurring in the southeastern part of the region under most scenarios. As for ozone, its concentration is projected to increase during 2016–35 under all emissions scenarios, compared to the baseline periods. The temporal evolutions of PM2.5 and ozone show PM2.5 reaching a peak during 2020–40, while ozone will likely increase steadily in the future.

基于Faster R-CNN的密集人群检测算法

为提高拥挤场景下的人群检测准确率,提出一种基于改进Faster R-CNN的密集人群检测算法。首先,在特征提取阶段添加空间与通道注意力机制,使用加强的双向特征金字塔网络(S-BiFPN)替代原网络中的多尺度特征金字塔(FPN),使网络对重要特征进行自主学习并加强对图像深层特征的提取;其次,引入多实例预测(MIP)算法对实例进行预测,以避免模型对拥挤场景下的目标造成漏检;最后,对模型中的非极大值抑制(NMS)进行优化,并额外增设一个交并比(IoU)阈值,以对检测结果的干扰项进行精确抑制。在开源的密集人群检测数据集上进行测试的结果显示,相较于原Faster R-CNN算法,所提算法的平均精度(AP)提升5.6%,Jaccard指数值提升3.2%。所提算法具有较高检测精度和稳定性,可以满足密集场景人群检测的需求。

Mutations around interferon sensitivity-determining region:A pilot resistance report of hepatitis C virus 1b in a Hong Kong population

AIM: To explore mutations around the interferon sensitivity-determining region (ISDR) which are associated with the resistance of hepatitis C virus 1b (HCV-1b) to interferon-α treatment. METHODS: Thirty-seven HCV-1b samples were ob- tained from Hong Kong patients who had completed the combined interferon-α/ribavirin treatment for more than one year with available response data. Nineteen of them were sustained virological responders, while 18 were non-responders. The amino acid sequences of the extended ISDR (eISDR) covering 64 amino acids upstream and 67 amino acids downstream from the previously reported ISDR were analyzed. RESULTS: One amino acid variation (I2268V, P = 0.023) was significantly correlated with treatment outcomein this pilot study with a limited number of patients, while two amino acid variations (R2260H, P = 0.05 and S2278T, P = 0.05) were weakly associated with treatment outcome. The extent of amino acid variations within the ISDR or eISDR was not correlated with treatment outcome as previously reported. CONCLUSION: Three amino acid mutations near but outside of ISDR may associate with interferon treatment resistance of HCV-1b patients in Hong Kong.

基于改进Mask R-CNN的多片烟叶部位的同步识别

为解决烟叶智能分级识别中需对多片散放烟叶同步进行部位识别的问题,提出一种基于改进Mask R-CNN的多片烟叶的部位同步识别方法:在Mask R-CNN区域建议网络中引入K-means聚类算法,对已标注目标检测框进行聚类,实现对预设的5种尺度的锚点尺寸和3种比例的锚点长宽比的优化,使其更加符合烟叶图像数据的分布特性,达到提高生成建议框的精确性、缩短识别时间的目的。基于采集的烟叶图像数据集,验证改进Mask R-CNN方法的有效性。结果表明,当IoU为0.5时,改进Mask R-CNN单样本耗时313 ms,比Mask R-CNN的326 ms快,在测试集上的均值平均精度(mAP)提高了3.56%。与Faster R-CNN和SSD目标检测算法相比,在准确率和召回率上也表现出优势。

基于改进Faster R-CNN算法的行人识别系统设计与研究

文章基于改进更快的区域卷积神经网络(Faster Region Convolutional Neural Networks,Faster R-CNN)模型,提出了一种行人识别系统设计。介绍了计算机视觉常用技术手段与方法、通行检测步骤,分析了主流的算法优缺点,利用深度学习方法提取图像特征,然后使用改进Faster R-CNN模型进行目标检测。在改进Faster R-CNN模型中,采用了自适应尺度池化和增强的感兴趣区域(Region of Interest,RoI)池化技术,可以提高模型检测精度和速度。

R/S AND WAVELET ANALYSIS ON EVOLUTIONARYPROCESS OF REGIONAL ECONOMIC DISPARITY IN CHINA DURING PAST 50 YEARS

This paper shows the dynamic process of regional disparity of economic development in China in the past 50 years from a new insight by using the rescaled range statistic (R/S) analysis and wavelet analysis of the Theil index sequence with different time scales. The main conclusions are: 1) The regional disparity of economic development in China, including the inter-provincial disparity, inter-regional disparity and intra-regional disparity, has existed for many years. Theil index by the comparative price has revealed the true trend for comparative disparity of regional economic development from 1952 to 2000. 2) Decomposition of Theil index indicates that the dynamic trend of comparative inter-provincial disparity in the coastal region is in line with dynamic trend of inter-provincial disparity in the whole China. 3) The R/S analysis results tell us that during 1966-1978, the Hurst exponent H=0.504≈0.5, which indicates that in that period the evolution of comparative inter-provincial disparity of economic development showed a random characteristic, and in the other periods, i.e. 1952-1965, 1979-1990 and 1991-2000, the Hurst exponent H>0.5, which indicates that in those periods the evolution of the comparative inter-provincial disparity of economic development in China had a long-enduring characteristic. 4) By using wavelet analysis at different time scale, we arrived at a conclusion that the evolutionary process of the disparity of economic development of China is not a simple inverted U shape but a compound of several U shapes. The result tells us that the evolutionary plot of inter-provincial disparity in China follows the inverted U on the whole at the higher scale, 24(16 years). That is to say, the disparity tends to rise in the first stage of economic development, and fall slowly over the peak in the second stage of economic development. However, if we shorten the time scale to 23 (8 years), then a link of several U shapes will appear.

FORWARD GATED-DIODE R-G CURRENT METHOD: A SIMPLE NOVEL TECHNIQUE FOR CHARACTERIZING LATERAL LIGHTLY DOPING REGION OF LDD MOSFET's

This paper presents a simple novel technique-forward gated-diode R-G current method-to determine the lateral lightly-doped source/drain (S/D) region interface state density and effective surface doping concentration of the lightly-doped drain (LDD) N- MOSFET's simultaneously. One interesting result of the numerical analysis is the direct characterization of the interface state density and characteristic gate voltage values corresponding to LDD effective surface doping concentration. It is observed that the S/D N- surface doping concentration and corresponding region's interface state density are R-G current peak position and amplitude dependent, respectively. It is convincible that the proposed method is well suitable for the characterization of deep sub-micron MOSFET's in the current ULSI technology.

Assessing Post-Conflict Challenges and Opportunities of the Animal-Agriculture System in the Alpine Region of Uvira District in Sud-Kivu Province, D. R. Congo

The Alpine region of Uvira District in South-Kivu Province faces low agricultural productivity challenges largely due to human population pressure on the natural resource base. During the dry season, conflicts between crop farmers and livestock breeders are common. This region is located 2500 - 3020 m a.s.l. and is almost inaccessible due to poor road infrastructure. To understand the interactions between natural resources (forests, pastures, soils) and sedentary agriculture, and to propose solutions for sustainable natural resource management, Participatory Rural Appraisal (PRA) sessions covering three farmer group interviews were conducted in July 2010 in Marungu, Kitembe and Kahololo location. The livelihoods of the Banyamulenge have been transformed from pastoralism to a sedentary system with the introduction of crop cultivation after 1980. Maize, bush beans and Irish potatoes are cultivated with limited success because of late crop maturity under the prevailing climatic conditions. Livestock, mainly cattle, sheep and goats is managed under extensive system, relying on natural pasture far from the human settlements. Livestock plays an important role to define people’s wealth and is mainly managed by men. There is a real need of forages during the dry season;however, cultivation of forages is unknown. On the other hand, pastures appear to be degenerating due to overgrazing. Access by farmers to efficient and effective agricultural extension services is fundamental in order to promote locally adapted and profitable crop and livestock production while at the same time, sustaining the productive resource base and security.

Assessment of Various Empirical Soil Loss Estimation Equations in Arid Regions

One of the most commonly used equations to estimate soil erosion is the revised universal soil loss equation (RUSLE). Based on the early approach developed by the Soil Conservation Service of USA, the rainfall erosivity factor (R-factor) in the RUSLE equation requires sub-daily rainfall data, which is usually not available. Other empirical equations estimate R-factor based on available rainfall data like annual and monthly rainfall data. In arid regions such as the Arabian Peninsula, several studies estimated the R-factor based on these empirical equations without calibration. We propose in this paper to assess the applicability of some of these empirical equations against R-factor values calculated using as a reference the RUSLE approach. For this data, data from 104 stations with sub-daily rainfall was collected. The reference R-factor was calculated for the 104 stations. The results of seven empirical equations were tested against the reference R-factor. Most of the tested equations significantly underestimated the R-factor. Furthermore, the obtained RMSE and MAE values were almost as high as the average R-factor, with MAPE exceeding 100%. Therefore, it is recommended not to apply these equations in arid regions. A recalibration of the form of equation that gave the best results, gave an RMSE of 280 (Mj·mm/(ha·hr)) and the MAPE dropped to 47.6%.

基于Faster R-CNN的人脸面部情感识别方法

常规人脸面部情感识别方法不准确,存在识别后的情感反馈误差大的问题,为此提出基于更快的区域卷积神经网络(Faster Region-Convolutional Neural Network,Faster R-CNN)的人脸面部情感识别方法。首先,采集人脸图像数据,通过面部检测、面部对齐、面部数据增强、面部归一化4个步骤预处理面部图像;其次,基于多尺度特征融合算法提取表情特征,生成情感识别数据标签;最后,利用FasterR-CNN构建人脸面部情感识别模型,并识别人脸面部情感。实验结果表明,基于FasterR-CNN的人脸面部情感识别方法在6种基本表情中均具有90%以上的识别准确率。

Relation between the evolution of seismic apparent strain field and the region of strong earthquake occurrence

In this paper, according to the data on the middle and strong earthquakes in China, we have preliminary studied the relation between the characteristic of space-time evolution of the seismic apparent strain field and the regions of 31 macroseism events since 1955. The result shows that, there is a rather well correlation between the anomaly region of seismic apparent strain and the zone of macroseism event occurrence within the time range of one to about five years. The R value of the application of the abnormal region of seismic apparent strain to predicting the area of strong earthquake occurrence is 0.458, and the empirical possibility of forecasting the region of macroseism occurrence is 0.625, and so the forecasting effect is comparatively well. Finally, the main results obtained above are discussed preliminarily.

基于R-FCN区域全卷积网络的绝缘子红外图像识别研究

红外热成像因其具有非接触性、灵敏性等优点,已被广泛应用于电力设备的带电检测及其诊断中。其中,对设备快速精确地识别定位是电力设备智能诊断的关键。然而利用传统机器算法对电力设备图像进行识别定位,存在泛化能力不强、鲁棒性较差等不足。针对此问题,开展基于R-FCN区域全卷积网络的绝缘子红外图像识别研究。在TensorFlow框架下搭建R-FCN检测模型,并利用迁移学习方法初始化模型权重,以提高训练效果。最后,将所研究算法与Faster-RCNN和SSD模型进行对比。实验表明,R-FCN模型的检测精度为89.2%,检测速度为23 fps,具有较高的精度和速度。该算法为绝缘子的智能诊断奠定坚实基础。

基于改进的FasterR-CNN苹果缺陷区域目标检测

目的:为实现在复杂果园环境下,初摘苹果快速、准确的分拣,基于Faster R-CNN模型,提出一种识别苹果损伤区域的算法模型。方法:修改Faster R-CNN模型特征提取网络,以苹果损坏区域为研究对象,将缺陷区域从图像中快速检测出来。结果:应用改进模型,对苹果缺陷区域检测的平均精度值较高(87.06%)。结论:应用改进的Faster R-CNN算法可以实现复杂环境下苹果较小缺陷区域快速检测,为果园苹果快速粗分拣设备开发提供技术支撑。

基于Faster R-CNN的海底管道智能检测方法

为提高海底管道缺陷及组件的检测精度并实现智能化海底管道安全检测,提出一种基于快速区域卷积神经网络(Faster R-CNN)的海底管道智能检测方法。首先,通过基值校正和分段映射-伪彩色化方法,将漏磁检测信号转化为伪彩色图,以增强漏磁信号的关键特征;其次,基于多模态数据增强来提升检测模型的泛化能力;然后,基于多模态数据增强后的样本训练改进的Faster R-CNN网络,建立最优的智能检测模型;最后,以试验场和渤海在役管道为例,验证所提方法的有效性。结果表明:所提方法的平均检测精度可达93.8%,相较原始的Faster R-CNN算法提高8%,且平均交并比达到0.75,能够精准地实现海底油气管道多目标检测,保障海底管道的安全运行。