您好,欢迎访问三七文档
当前位置:首页 > 电子/通信 > 数据通信与网络 > 信息融合技术Lecture4
Lecture 4: Common Representational Format 周鑫xzhou@nuaa.edu.cnIntroduction•Convert all sensor observations into a common format is a basic requirementSpatialAlignmentTemporalAlignmentSensorValueNormalizationImagecourtesyofmPower3/EmergeImagecourtesyofMassachusettsExecutiveOfficeofEnvironmentalAffairsI1I2Kriging•Convert a set of sensor measurements•Krigingrefers to a family of best linear unbiased estimators•, are measurements located in the neighborhood of a given point the estimated value at is into a continuous map. GISKriging(contd.)Kriging(contd.)•Page 56, line 7•Should beCommon Representational Format•When choose a common representational format, we consider these factors–Completeness. Support a complete and non‐redundant representation–Uncertainty Representation. Support an explicit description of uncertainty in the environment–Computation and Storage. Support efficient computations and have low storage requirements–Adaptability. Capable of adapting itself to a wide variety of circumstances and applications.Example •Representation of Spatial Environment for Mobile Robots. Topological Representation and Occupancy Grid Representation.Subspace Methods•Reduce the dimension of the raw data. Important information is not lost, keep the computational load and/or the storage requirements low.•Principal Component Analysis (PCA)•Linear DiscriminantAnalysis (LDA)Principal Component Analysis•Input data•A reduced L‐dimensional space, L‹M,defined by•is: Principal Component Analysis (contd.)•Example (matlab)Linear DiscriminantAnalysis•Suppose each input measurement is associated with a given class •LDA is to find a L‐dimensional subspace in which the different classes are maximally separated.Linear DiscriminantAnalysis(contd.)Multiple Training Sets•Ensemble learning –An ensemble (collection) of functions or classifiers Sm–Each Smis learnt on its own training set Dm–Generate training sets Dmfrom a common training set D–Sub‐sampling •Cross‐Validation•Bagging (Bootstrap)•Wagging•BoostingBootstrappingExample,Randomlysamplingwithreplacementwithauniformselection
本文标题:信息融合技术Lecture4
链接地址:https://www.777doc.com/doc-46171 .html