福建省复杂地形下旬平均气温分布式模拟

为建立高时空分辨率的福建省复杂地形下气温栅格数据集,利用福建省及其周边33个常规气象站观测资料,基于数字高程模型（DEM）数据,综合考虑海拔、太阳总辐射、地表长波有效辐射对旬平均气温的影响,模拟了福建省复杂地形下旬均温的空间分布。结果表明：1）常规站验证结果显示：各旬气温绝对误差平均值（MAE）最小为0.46℃,最大为2.3℃,全年平均为0.87℃;加密站验证结果显示,MAE最大为2.3℃,最小0.5℃,全年平均为0.96℃。2）模拟结果能反映旬均温的宏观分布规律与局地细节特征。宏观范围内,旬均温受纬度影响较大,由北至南气温逐渐升高,沿海地区旬均温整体高于内陆,山区旬均温明显较低;局地范围内,各坡向上气温差异显著,海拔越高、坡度越大,差异越明显;地形因子对旬平均温的影响具有季节差异,具体表现为冬季时地形因子对旬均温的影响最大,秋季次之,春夏季节中地形因子对旬均温的影响最弱。     

基于卷积神经网络与条件随机场方法提取乡镇非正规固体废弃物

随着村镇经济建设发展,生活垃圾和工业固体废弃物造成的污染问题日益突出,已经成为制约新农村建设发展和生态文明建设的关键问题,而目前针对乡镇非正规固体废弃物的调查与统计主要依赖全国各乡镇相关部门逐级调查上报,工作量较大。本文基于高分辨率遥感影像,将深度学习模型和条件随机场模型相结合引入到乡镇固体废弃物的提取研究中,探索一种基于深度卷积神经网络的乡镇固体废弃物提取模型。由于固体废弃物在影像上表现为面积小,分布破碎等特点,为了提高工作效率,将模型特分为识别和提取2个部分：① 通过全连接卷积网络（CNN）对固体废弃物进行快速识别判断,筛选感兴趣区域影像块;② 在传统的全卷积神经网络(FCN)的基础上加入条件随机场模型(CRF)提取固体废弃物边界,提高整体分割精度。根据安徽、山西等地区相关部门上报固体废弃物堆放点以及住房与城乡建设部城乡规划管理中心进行野外检查的结果,实验最终识别精度达到86.87%以上;形状提取精度为89.84%,Kappa系数为0.7851,识别与提取精度均优于传统分类方法。同时,该方法已经逐步应用于住房和城乡建设部有关成都、兰州、河北等部分乡镇非正规固体废弃物的核查工作,取得了较为满意的结果。     

基于随机森林的光子计数激光雷达点云滤波

新一代星载激光雷达卫星ICESat-2首次采用了微脉冲光子计数激光雷达技术,由于单光子探测的灵敏性导致数据在大气和地表下层产生了大量噪声,因此对光子计数激光雷达点云数据实现信号和噪声的分离是开展进一步应用研究的前提和基础。本文选择美国俄勒冈州和弗吉尼亚州2个研究区,采用MATLAS数据,根据光子点云数据的特点构造了12个光子点云特征,对所构造的特征利用随机森林进行变量筛选,用机器学习方法对光子点云进行分类,并将建立好的模型推广到整个研究区。研究结果表明,本文构建的分类器分类总精度达到了96.79%,Kappa系数为0.94,平均生产者精度和用户精度分别为97.1%和96.8%。在相对弱噪声、平坦地形区域和强噪声、复杂地形区域都取得较好的分类结果。本文结果显示了基于少量样本通过机器学习的方法构建模型,可以推广到较大范围区域的光子点云分类应用中。     

无人机倾斜影像自动检索及影像姿态恢复

无人机倾斜摄影测量颠覆了以往正射影像只能从垂直角度拍摄的局限,在三维建模中有广泛的前景。针对有些无人机倾斜影像数据无相机标定参数、无航带信息（无序）、无POS信息的现状,本文以计算机视觉中基于内容的影像检索方法与改进的渐进式SFM方法为基础,提出一种“三无”影像自动检索、空中三角测量及影像三维重建的方法。该方法首先通过提取的特征检索出相似影像并建立网络结构,然后将影像进行两两匹配增强对应关系并进行连接点的追踪,最后利用光束法平差方法对其进行平差,获取影像集的三维点云,提高大规模影像检索、影像匹配速度的同时,提高重建的精确性和鲁棒度。本文选取三组典型试验区大数据量倾斜影像数据进行试验,立体实测控制点中误差可以达到平面0.16m/高程0.18m,试验验证了方法的稳定性、可靠性和实用性。     

基于动态极坐标参数化的遥感影像匹配方法

图像匹配作为三维重建至关重要的环节,其精度直接影响了平差优化、正射校正等模块的精度。对于城镇、农场等特征密集型区域,特征距离小,相似性强,易于匹配图像;而针对草地、沙漠等特征不明显区域,特征距离大,如果使用特征点匹配的方法,严格阈值下难以获得足够数量的匹配对,放宽阈值又将引入较多误匹配对,这也是导致稀疏点云不够均匀的原因之一。在此场景下,本文提出了基于动态极坐标参数化的无人机正视影像匹配算法,首先对图像做极坐标参数变化,采用动态策略解决极轴方向采样不均匀的问题,使用最小二乘法对得到的极坐标影像对做位移方向上的匹配,匹配后得到的旋转量和平移量,将该结果和SIFT算法的结果做比较。本文设计了2组实验,即参数已知的解算实验和参数未知的解算实验,且每组实验进行3次。在同等配置的计算机上,对两张7360像素×5400像素,32位的影像,本文方法的位姿解算时间相比SIFT的时间减少约57%,二者求得的位姿差通常小于1%。结论表明二者的结果在精度上表现相当,在时间上明显优于SIFT算法,具有实际的应用价值。     

基于深度学习的钨钼找矿靶区预测方法研究

随着矿产勘查工作由浅部矿向深部隐伏矿、由易识别矿向难识别矿发展,找矿难度日益增大,地质专家越来越重视新理论、新方法、新技术的应用。深度学习作为人工智能的前沿领域/技术,对于实现矿产资源预测“智能化预测评价”具有得天独厚的优势。本文以陕西省镇安县西部钨钼矿集区单元素化探异常原始数据为基础,提出了基于深度学习的钨钼矿产评价方法。该方法以归一化地球化学数据作为模型训练数据,通过深度学习中深度自编码网络方法实现异常值提取进而识别重点成矿有利地段,实现矿产资源找矿远景区定性预测。研究结果表明,在对957条单元素化探异常原始数据分类且做好模型标签后,整个过程在计算机的“黑盒子”中自动完成学习和预测,相较于传统预测研究方法,本文方法具有自动化程度高和客观性强的特征。此外,本文利用已知矿点构建训练数据集,采用随机森林方法对预测区进行矿产资源找矿靶区预测圈定,为进一步缩小找矿靶区范围提供科学依据。     

Comparisons of several evaporation/precipitation datasets for the Bohai Sea based on salinity simulation

Evaporation (E) rate and precipitation (P) rate are two significant meteorological elements required in the ocean baroclinic modeling as external forcings. However, there are some uncertainties in the currently used E/P rates datasets, especially in terms of the data quality. In this study, we collected E/P rates data from ERA-40, NCEP/NCAR Reanalysis, HOAPS for the Bohai Sea and nine routine stations around Laizhou Bay, and made comparisons among them. It was found that the differences in E/P rates between land and sea are remarkable, which was due to the difference in underlying surfaces. Therefore, the traditional way of using E/P rates acquired on land directly at sea is not correct. Since no final conclusion has been reached concerning the net water transport between the Bohai Sea and the Yellow Sea, it is unfeasible to judge the adequacy of the four kinds of data by using the water budget equation. However, the E/P rates at ERA-40 sea points were considered to be the optimal in terms of temporal/spatial coverage and resolution for the hindcast of salinity variation in the Bohai Sea. Besides, using the 3-D hydrodynamic model HAMSOM (HAMburg Shelf Ocean Model), we performed numerical experiments with different E/P datasets and found that the E/P rates at sea points from ERA-40 dataset are better than those from NCEP/NCAR Reanalysis dataset. If NCEP/NCAR Reanalysis E/P rates are to be used, they need to be adjusted and tested prior to simulation so that more close-to-reality salinity values can be reproduced.     

Radar cross section analysis of marine targets using a combining method of physical optics/geometric optics and a Monte-Carlo simulation

In this paper, the radar cross section of flat plates on ocean surfaces is statistically investigated. A combining method of physical optics and geometric optics is applied to establish an effective backscattering analysis procedure. This method is a high-frequency analysis method originally derived from a simplified Stratton-Chu integral equation, assuming that the radar is far away from the target so that Kirchhoff approximation is valid. A Monte-Carlo simulation method is adopted to statistically analyze the effects of undulated ocean surfaces. The ocean surfaces are randomly generated by Pierson-Moskowitz ocean wave spectrum and a directional distribution function. Numerical investigations are carried out for flat plates, with the same height and width but with different inclined angles, on ocean surfaces of various significant wave heights.     

Aequorea taiwanensis n. sp. (Hydrozoa, Leptomedusae) and mtCOI sequence analysis for the genus Aequorea

Aequorea taiwanensis, a new hydrozoan species from the Taiwan Strait was described using morphological and molecular characteristics. Both morphological and mitochondrial cytochrome oxidase subunit I (mtCOI) data supported A. taiwanensis n. sp. as a valid species. Sequence divergence and genetic distance of A. taiwanensis n. sp., A. papillata and A. conica were analysed based on the mtCOI gene sequences. The mtCOI sequences from these three species of the genus Aequorea showed high variation frequency, with sequence divergences ranging from 9.10% to 11.9%, and pairwise genetic distances ranging from 0.097 to 0.130. MtCOI sequence analysis provided diagnostic molecular systematic characteristics for accurate identification and discrimination of the species of Aequorea or their populations, and will be used to resolve evolutionary relationships among them.It was suggested that 10%—20% pairwise mtCOI sequence differences indicated the specieslevel divergence among congeneric species in the Hydromedusae.     

Nonlinear numerical simulation on extreme-wave kinematics

A fully nonlinear numerical model based on a time-domain higher-order boundary element method (HOBEM) is founded to simulate the kinematics of extreme waves. In the model, the fully nonlinear free surface boundary conditions are satisfied and a semi-mixed Euler-Lagrange method is used to track free surface; a fourth-order Runga-Kutta technique is adopted to refresh the wave elevation and velocity potential on the free surface at each time step; an image Green function is used in the numerical wave tank so that the integrations on the lateral surfaces and bottom are excluded. The extreme waves are generated by the method of wave focusing. The physical experiments are carried out in a wave flume. On the horizontal velocity of the measured point, numerical solutions agree well with experimental results. The characteristics of the nonlinear extreme-wave kinematics and the velocity distribution are studied here.