类间数据不均衡条件下基于平衡随机森林的轴向柱塞泵故障诊断方法

针对轴向柱塞泵实际故障诊断中采集到的故障类数据远少于正常类数据的情况,为提升故障分类精确率,提出了一种基于平衡随机森林(Balanced Random Forest,BRF)的轴向柱塞泵故障诊断方法.BRF算法是随机森林(Random Forest,RF)的改进算法,将欠采样方法与RF结合,强化了RF处理非均衡数据的能...     

结合C/S架构和BRF算法的移动增强现实研究

目的提出一种结合C/S(Client/Server)架构和BRF(Boosted random ferns)算法的移动增强现实应用方案,以保证图像识别算法对于产品外包装的识别性能。方法 BRF是一种高效、鲁棒的特征匹配算法,但由于手机内存及处理器等硬件条件的制约,不能直接适用于手机终端。将C/S模式与BRF算法相结合应用于图像特征匹配,并设计实验测试比较文中方案(CS-BRF)与ORB算法的识别速度和匹配精度。结果实验结果表明,相比ORB算法,CS-BRF在识别速度相近的前提下,具有更为优异的识别精度。结论 CS-BRF能够实时准确识别印刷品图像,良好适用于产品包装移动增强现实系统。     

Evaluation of the consistency of the three MRPV model parameters provided by the MISR level 2 land surface products: a case study in Mainland Spain

The multiangular Rahman–Pinty–Verstraete modified (MRPV) semi-empirical model uses three parameters (ρ₀, Θ, and k) for describing the anisotropy of an arbitrary target. They have been usefully proved to characterize some forest attributes and land covers. However, there is no enough evaluation of the consistency of this product, and the possible affection from different factors in the reliability of them. Here, we explored the consistency of the MRPV parameters provided in the MISR L2 Land Surface (MIL2ASLS) product, with data from Mainland Spain, grouping MISR images into close time pairs. Thus, it was studied the three MRPV parameters through retrieving Spearman’s rank correlation coefficient (ρ) and mean relative differences related to every pair of images. The results showed the ρ₀ parameter presented higher consistency than the others, with ρ over 0.85 and meant relative differences around 15%. The k parameter showed ρ over 0.65 and average relative disagreements over 8%. Finally, the Θ parameter reached ρ around 0.60. The Θ mean differences were over 25% unless the combination of the blue band which was especially bad and its values were up to 50%. So, it is crucial having into account when the parameters of this product are used to look into the band and the own parameter.     

基于RBF网络的粮食水分检测数据融合研究

为了提高测量的准确性和快捷性,需要融合处理多传感器检测的数据。本文首先介绍BRF网络的特性和训练方式,然后进行样本数据采集、样本数据归一化、神经网络的训练及其结构的确定,完成基于RBF网络的水分检测数据处理过程,实现粮食水分检测中的多传感数据融合。经过Matlab中的神经网络模型训练后,实验结果表明,拟合值始终在目标值上下波动,波动的范围在7%以内,该方法具有较大的优越性,可在其它工业领域中推广应用。     

野外目标BRF数据处理方法初探

通过对野外目标方向反射数据的处理可以获得其BRF特性,讨论了用数据优化、归一化、实时参数修正、绘图、比较等对野外目标BRF数据进行处理的初步方法,并用这些方法对1996年9月敦煌试验场的测量数据作了处理,分析和检验了处理结果。另外对BRF数据方法规范化也作了探讨。     

Using Radial-basis Function Network for CLV

Analysis and comparing with three existing and popularly used forcasting customer lifetime value(CLV) methods,which are the Dwyer method,customer event-method and fitting method,and using performances of the existent artificial neural network,we apply the Radial-basis Function (RBF) network to forecast the CLV,the RBF network can approach the objective function partially.To every input/output pairs,it only needs adjust the weight a little and learn quickly which is very important to the forecast precision.Simulation and experimental results on the customers‘ data of a company in Shaanxi Province reveal the effectiveness and feasibility of the RBF network.     

电锅炉温度控制系统PID参数整定算法的改进

针对电锅炉温度控制系统常规PID控制性能欠佳的问题,提出一种PID参数整定的改进算法。先用RBF网络在线辨识温度采样信息,得到控制器所需的Jacobian信息,然后用LM算法代替传统的梯度法整定PID参数,得到改进的PID算法。通过电锅炉温度控制系统仿真分析,验证了算法的快速性、抗扰性、鲁棒性,具有较好的应用前景。     

Detection of foliage conditions and disturbance from multi-angular high spectral resolution remote sensing

Disturbance of forest ecosystems, an important component of the terrestrial carbon cycle, has become a focus of research over recent years, as global warming is about to increase the frequency and severity of natural disturbance events. Remote sensing offers unique opportunities for detection of forest disturbance at multiple scales; however, spatially and temporally continuous mapping of non-stand replacing disturbance remains challenging. First, most high spatial resolution satellite sensors have relatively broad spectral ranges with bandwidths unsuitable for detection of subtle, stress induced, features in canopy reflectance. Second, directional and background reflectance effects, induced by the interactions between the sun-sensor geometry and the observed canopy surface, make up-scaling of empirically derived relationships between changes in spectral reflectance and vegetation conditions difficult. Using an automated tower based spectroradiometer, we analyse the interactions between canopy level reflectance and different stages of disturbance occurring in a mountain pine beetle infested lodgepole pine stand in northern interior British Columbia, Canada, during the 2007 growing season. Directional reflectance effects were modelled using a bidirectional reflectance distribution function (BRDF) acquired from high frequency multi-angular spectral observations. Key wavebands for observing changes in directionally corrected canopy spectra were identified using discriminant analysis and highly significant correlations between canopy reflectance and field measured disturbance levels were found for several broad and narrow waveband vegetation indices (for instance, r²_NDVI = 0.90; r²_CHL3 = 0.85; p < 0.05). Results indicate that multi-angular observations are useful for extraction of disturbance related changes in canopy reflectance, in particular the temporally and spectrally dense data detected changes in chlorophyll content well. This study will help guide and inform future efforts to map forest health conditions at landscape and over increasingly coarse scales.     

蓝牙模块和OMAP5910的接口技术

研究一种嵌入式数字信号处理器（DSP）OMAP5910的蓝牙接口实现。整个系统以DSP为核心,采用TI公司的BRF6100蓝牙模块实现V1．1协议,进行无线信号收发;DSP配置音频AD／DA。实现音频信号采集和处理。该系统制造简单,集成度高,适用于各种家庭和个人无线设备。     

AERONET-based surface reflectance validation network (ASRVN) data evaluation: Case study for railroad valley calibration site

The AERONET-based Surface Reflectance Validation Network (ASRVN) is an operational processing system developed for validation of satellite derived surface reflectance products at regional and global scales. The ASRVN receives 50 × 50 km² subsets of MODIS data centered at AERONET sites along with AERONET aerosol and water vapor data, and performs an atmospheric correction. The ASRVN produces surface bidirectional reflectance factor (BRF), albedo, parameters of the Ross-Thick Li-Sparse (RTLS) BRF model, as well as Hemispherical-Directional Reflectance Factor (HDRF), which is required for comparison with the ground-based measurements. This paper presents a comparison of ASRVN HDRF with the ground-based HDRF measurements collected during 2001-2008 over a bright calibration Railroad Valley, Nevada site as part of the MODIS land validation program. The ground measurements were conducted by the Remote Sensing Group (RSG) at the University of Arizona using an ASD spectrometer. The study reveals a good agreement between ASRVN and RSG HDRF for both MODIS Terra and Aqua with rmse ~ 0.01-0.025 in the 500 m MODIS land bands B1-B7. Obtained rmse is below uncertainties due to the spatial and seasonal variability of the bright calibration 1 km² area. While two MODIS instruments have a similar rmse in the visible bands, MODIS Aqua has a better agreement (lower rmse) with the ground data than MODIS Terra at wavelengths 0.87-2.1 μm. An independent overall good agreement of two MODIS instruments with the ground data indicates that the relative calibration of MODIS Terra and Aqua at medium-to-bright reflectance levels for the stated time period is significantly better than uncertainties of the ASRVN and ground data.