基于信息融合的农业自主车辆障碍物检测方法

针对单一传感器在智能车辆环境感知中的局限性,提出一种基于摄像机与激光雷达信息融合的农业自主车辆前方障碍物检测方法。对单目摄像机获取的图像进行基于Ft（Frequency-tuned）算法的显著性检测,并生成显著图。同时对激光雷达反射点进行基于数据关联性评估的聚类分析,确定障碍物数量、边界与位置等先验信息。然后以激光雷达坐标相对应的图像像素坐标为种子点,由种子点激活经过处理的显著图,基于受限区域生长实现障碍物区域分割。试验结果表明,基于Ft算法的图像显著性检测具有更好的边缘检测效果,基于种子点的受限区域生长法可以有效地进行障碍物分割。在机器视觉的基础上融入激光雷达数据,可以更好地排除非障碍物的干扰,实现了障碍物的完整检出。     

基于无人机遥感与卷积神经网络的草原物种分类方法

基于无人机高光谱成像遥感系统,在400～1 000 nm波段内采集低矮、混杂生长的荒漠草原退化指示物种的高光谱图像信息。分别在退化指示物种的开花期、结实期和黄枯期进行飞行实验,飞行高度30 m,高光谱图像地面分辨率2. 3 cm。采用特征波段提取与深度学习卷积神经网络相结合的方式,提出一种荒漠草原物种水平分类的方法,结合植物物候给出了中国内蒙古中部荒漠草原物种分类的推荐时相,总体分类精度和Kappa系数平均值分别达到94%和0. 91。研究结果表明,无人机高光谱成像遥感技术及深度卷积神经网络可以较好地实现荒漠草原退化指示物种的分类,与基于径向基核函数的支持向量机、基于主成分分析的深度卷积神经网络分类法相比,基于特征波段选择的深度卷积神经网络分类法效果最好,分类精度最高。无人机搭载高光谱成像仪低空遥感和卷积神经网络法提供了一种草原物种水平分类的途径。     

小麦倒伏信息无人机多时相遥感提取方法

采用两期无人机可见光遥感图像,对灌浆期冬小麦倒伏图像特征及倒伏信息提取方法进行研究。从增强图像空间域方面,对图像进行二次低通滤波,获取地物散点图,以散点存在明显分界线为判定标准,选出小麦倒伏信息提取的单特征,对两单特征线性拟合构建倒伏小麦两时期提取特征参数F_1和F_2,再以两特征参数相似性构建综合特征参数F_3。将特征参数结合K-means算法提取冬小麦倒伏信息,整体精度(OA)达86. 44%以上,Kappa系数达0. 73以上,倒伏信息提取精度(F)为81. 07%以上,因此综合特征参数可作为两个时期冬小麦倒伏信息提取特征参数。分别用本文方法、支持向量机、神经网络法和最大似然法提取验证区域倒伏小麦信息,经验证,本文方法提取小麦倒伏信息整体精度(OA)达86. 29%以上,Kappa系数达0. 71以上,倒伏信息提取精度(F)达80. 60%以上;其他3种常用方法提取的整体精度(OA)为69. 68%～87. 44%,Kappa系数为0. 49～0. 72,倒伏信息提取精度(F)为65. 33%～79. 76%。结果表明,本文方法整体精度和倒伏信息提取精度均高于目前常用分类方法。因此,综合特征参数与K-means算法对冬小麦在灌浆期倒伏信息提取具有一定的准确性和适用性。     

车辆换挡系统调压阀优化设计与试验

针对车辆换挡充油过程中离合器充油压力曲线的实际值与理想值之间存在偏差的问题,建立了包含供油单元、电磁换向阀、调压阀和离合器的换挡系统仿真模型,采用AMESim对充油过程进行仿真分析。以充油压力为优化目标,应用误差积分准则确定优化函数,通过Pareto图得到调压阀中影响充油压力的关键参数为阀口开度、弹簧预紧力和阻尼孔直径。分别采用二次拉格朗日非线性规划算法(Non-linear programming by quadratic Lagrangian,NLPQL)和遗传算法(Genetic algorithms,GA)对上述参数进行优化。对比优化后的仿真充油曲线与理想充油曲线,可知遗传算法为最优算法,并确定了调压阀的最优结构参数。根据控制变量法设计试验方案,结果表明,采用遗传算法优化后,换挡系统的稳定性在一定程度上得到提高,其充油过程更加符合理想充油过程,换挡品质得到改善。     

拖拉机田间作业参数无线检测系统研究

针对当前拖拉机检测系统功能集成度低、检测参数不全面、传输距离有限的问题,开发了拖拉机田间作业参数无线检测系统。该系统由传感器、数据采集仪及上位机软件监测平台3部分组成,能够实现PTO转矩及转速、油耗、发动机转速、悬挂提升力、力位调节加载力、加载角度、行驶速度、车轮转速、牵引力等多种参数的采集、无线发送与存储。系统工作时,数据采集仪中的车载检测仪将采集的传感器数据发送至无线数据接收器,无线数据接收器通过串口将数据传输至上位机软件监测平台,实现对各类试验参数的实时监测与数据处理。为验证检测系统的可行性与稳定性,对系统进行了采集通道的计量,结果显示模拟信号通道绝对误差绝对值最大为0.003V,引用误差最大为0.03%,频率信号通道检测绝对误差最大为2Hz,引用误差最大为0.013%,满足对拖拉机作业参数的采集需求。在此基础上,进行了PTO转矩参数及拖拉机无负载行驶速度采集试验。试验结果表明,检测系统可以实现转矩参数的稳定采集及数据的无线传输;在5、8、14km/h 3挡车速匀速行驶下,拖拉机车轮转速与实际行驶速度基本一致,最大相对误差分别为2.0%、1.2%及0.7%。本系统可满足对拖拉机工作性能参数的无线检测需求,数据采集稳定且采集精度较高,为拖拉机多作业参数的无线采集提供有效手段。     

浅谈无人机遥感技术在土地整治项目中的应用

随着科学技术水平不断推进,计算机技术与遥感技术等有机结合,无人机遥感技术凭借其轻便、高效、精确的数字地形测绘技术被广泛应用于各类工程领域,本文结合土地整治项目对无人机遥感技术发展现状以及在土地整治项目中的应用优势和特点进行研究分析,并探讨了无人机遥感技术应用在土地整治项目中的优势和不足,无人机遥感技术虽能提供基础的数据资料和技术支持,但项目实施中仍要对部分区域进行必要的实地核查。     

基于无人机高光谱遥感影像的防护林树种分类

树种信息对林业资源监测和管理具有重要意义,及时准确地掌握树种及长势状况是防护林工程建设与效益评价的基础。为研究利用无人机高光谱数据进行防护林树种分类的效果,选取典型区域使用Matrice600型六旋翼无人机搭载Rikola高光谱成像仪获取高光谱影像,基于支持向量机-递归特征消除算法（SVM-RFE）选取原始波段最佳组合,再结合纹理特征、植被指数和数理统计特征,使用随机森林算法对所有特征进行重要性评估并与分类精度相结合进行特征优化,进而构建高光谱影像全波段、原始波段最佳组合、全部特征变量、基于随机森林（RF）特征优化后特征变量4种分类方案,分别采用最大似然法（MLC）、支持向量机（SVM）、随机森林对防护林优势树种进行分类。结果表明：所提出的基于交叉验证的SVM-RFE算法选出的原始波段组合能更好地还原原始光谱特征;通过RF算法的特征重要性分析与分类精度相结合的方法可以有效选出重要特征,当使用全部特征的85%（包括17个光谱特征、3个纹理特征、5个植被指数和3个数理统计特征）进行分类时,总体精度最高为9593%（Kappa系数为0.9475）;所有特征中植被指数特征最重要,3种分类方法中RF算法分类总体精度（OA）最高。     

The concept of habitat specificity revisited

Habitat specificity indices reflect richness (α) and/or distinctiveness (β) components of diversity. The latter may be defined by α and γ (landscape) diversity in two alternative ways: multiplicatively () and additively (). We demonstrate that the original habitat specificity concept of Wagner and Edwards (Landscape Ecol 16:121–131, 2001) consists of three independent components: core habitat specificity (uniqueness of the species composition), patch area and patch species richness. We describe habitat specificity as a family of indices that may include either area or richness components, or none or both, and open for use of different types of mean in calculation of core habitat specificity. Core habitat specificity is a beta diversity measure: the effective number of completely distinct communities in the landscape. Habitat specificity weighted by species number is a gamma diversity measure: the effective number of species that a patch contributes to landscape richness. We compared 12 habitat specificity indices by theoretical reasoning and by use of field data (vascular plant species in SE Norwegian agricultural landscapes). Habitat specificity indices are strongly influenced by weights for patch area and patch species richness, and the relative contribution of rare vs. common species (type of mean). The relevance of properties emphasized by each habitat specificity index for evaluation of patches in a biodiversity context is discussed. Core habitat specificity is emphasized as an ecologically interpretable measure that specifically addresses patch uniqueness while habitat specificity weighted by species number combines species richness and species composition in ways relevant for conservation biological assessment. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Modeling the water budget in a deep caldera lake and its hydrologic assessment: Lake Ikeda, Japan

The hydrologic assessment of a lake water budget can be helpful in achieving proper water management and sustainable water use. A model to analyze a lake water budget was developed and verified for Lake Ikeda, Japan. Lake evaporation was estimated by numerical analyses of lake water temperature and the lake energy budget. Inflow from the lake catchment area and leakage from the lake bottom were estimated based on the tank model and Darcy's law, and the model parameters were optimized by the shuffled complex evolution method. The estimated monthly lake evaporation rate is consistent with the evaporation rate estimated by the energy budget Bowen ratio method based on in situ data from 2004 to 2005. Moreover, the calculated time series of daily lake levels agrees well with those of measured lake levels during 1983 to 1999. Thus, the model is useful for evaluating the lake water budget. Numerical analysis reveals seasonal and annual variation characteristics in the water budget components. Precipitation, inflow from the catchment area, and river water supply are generally high during the rainy season from June to July with substantial annual variation. Lake evaporation is greatest in October and least in April, but the annual variation is relatively small. Agricultural water use is relatively high from April to September. There are no marked seasonal changes in leakage and drinking water use. The lake level is generally highest in September and lowest in March, which is characterized by seasonal changes in water budget components. The model was also applied to 17-year simulations under hypothetical hydrologic conditions to examine the effect of water use and agricultural water management on the lake level. Results indicate that river water supply, provided under the agricultural water management system, effectively compensates for the decrease in lake water resulting from agricultural water use.     

Combined use of optical and radar satellite data for the detection of tillage and irrigation operations: Case study in Central Morocco

The objective of this study is to present a new application of optical and radar remote sensing with high spatial (∼10 m) and temporal (a few days) resolutions for the detection of tillage and irrigation operations. The analysis was performed for irrigated wheat crops in the semi-arid Tensift/Marrakech plain (Central Morocco) using three FORMOSAT-2 images and two ASAR images acquired within one week at the beginning of the 2005/2006 agricultural season.The approach we developed uses simple mapping algorithms (band thresholding and decision tree) for the characterisation of soil surface states. The first images acquired by FORMOSAT and ASAR were processed to classify fields into three main categories: ploughed (in depth), prepared to be sown (harrowed), and not ploughed-not harrowed. This information was combined with a change detection analysis based on multitemporal images to identify harrowing and irrigation operations which occurred between two satellite observations.The performance of the algorithm was evaluated using data related to land use and agricultural practices collected on 124 fields. The analysis shows that drastic changes of surface states caused by ploughing or irrigation are detected without ambiguity (consistency index of 96%). This study provided evidence that optical and radar data contain complementary information for the detection of agricultural operations at the beginning of agricultural season. This information could be useful in regional decision support systems to refine crop calendars and to improve prediction of crop water needs over large areas.