Multi-source Data fusion Learning

Recently, I am looking at the paper on Multi-Source data fusion, and record the reading notes and some opinions of myself.

Multi-Source Information Fusion (Multi-Source information Fusion) was first proposed by American scholars, which is a new and overlapping field, which has been widely developed in recent years. It is used in many fields: target recognition, remote sensing, medicine and so on.
Multi-source information fusion is a recognition of a variety of data, in the process of synthesizing and judging, the data that participates in fusion often has: multi-source, heterogeneity, incompleteness, etc., according to different levels of fusion, information fusion can be divided into: data-level fusion, model-level fusion (feature-level fusion), decision-level fusion.
Data-level fusion is the most low-level fusion, directly to the original data processing, the advantage is to retain the original information, information loss is very small, the disadvantage is the fusion of the limitations of large, can only single or the same type of data processing, the calculation of a large amount.
Model-level fusion is in three kinds of fusion between the middle level, more intelligent, the advantage is the original data extraction and processing of the fusion, in the amount of data is reduced, the result is the reduction of computational volume, the disadvantage is that information loss will bring data precision drop. Decision-level fusion in the three is the highest level of integration, is the most intelligent fusion, is built on the basis of model integration for the final processing results of the comprehensive decision-making.
Something's going on. It can be used to fuse different types of data, with a small amount of computation, fault tolerance and anti-interference, but the disadvantage is obvious, the loss of data information will bring about a decrease in precision.     The three comparisons are as follows: data-level Fusion model-level Fusion decision-level Fusion information processing volume maximum small minimum information loss minimum maximum anti-interference ability         Least small minimum fault tolerant ability worst poor good fusion algorithm difficult to fusion before processing the smallest
The best fusion performance of the most of the difference sensor dependence degree of the key technologies include: Data conversion, data Association, Fusion algorithm.        Comparison and fusion method of common fusion methods running environment information type information expressing uncertainty fusion technology applicable range weighted average dynamic redundancy original value read weighted average   Dynamic redundancy probability distribution of low-level fusion Kalman Filter     Gaussian noise system model filtering low-level fusion Bayesian estimation static redundancy probability distribution Gaussian noise Bayesian estimation low fusion statistical decision static redundancy probability distribution high        The high-level fusion evidence theory of noise extremum decision-making the static redundancy complementary Proposition reasoning high-level fusion fuzzy theory static redundancy complementary proposition     Membership degree logic inference high-rise fusion neuron network static and dynamic redundancy complementary neuron input learning error neural network low or high generating rule static and dynamic redundancy complementary proposition confidence factor

























 Logic reasoning High Level fusion