There are many methods for sensor information fusion, but so far, the most commonly used methods mainly include three categories:Embedded constraint method, evidence combination method, artificial neural network method.

oneEmbedding constraint method

The embedding constraint method considers that there are multiplesensorThe objective environment obtained(The tested object)Multiple sets of data are images formed by the objective environment according to a certain mapping relationship, and information fusion is to solve the original image through images, that is, to understand the objective environment. Using mathematicsThat is to say, all the information from sensors can only describe certain aspects of the environment, and there are many environments with these characteristics. To make a set of data correspond to a unique environment(The above mapping is oneonemapping)It is necessary to impose constraints on the original image of the mapping and the mapping itself, so that the problem can have a unique solution.

twoCombination of evidence method

The combination of evidence method holds that completing a certain intelligent task is to make several possible decisions based on information about a certain aspect of the environment, and manysensorData information to some extent reflects the situation in this aspect of the environment. Therefore, we analyze each-The support degree of data as evidence supporting a certain decision, and the support degree of different sensor data is combined, that is, evidence combination, divided intoThe decision with the highest degree of combined evidence support is identified as the result of information fusion.

The combination of evidence method is aimed at completing a certain taskoneTo process data information from multiple sensors and complete an intelligent task as required by the taskIn fact, it is a decision to take a certain action. It starts with a single onesensorProvide a measure of the level of support for each possible decision based on data information(The degree to which data information serves as evidence to support decision-making)Then search for a method or rule to combine evidence,When two different sensor data are known(That is, evidence)When it comes to the degree of support for decision-making, repeatedly applying combination rules is the most effectiveThe overall level of support for a decision by the consortium of all data information. The decision that receives the greatest evidence support is the result of information fusion.

The key to using evidence combination for data fusion is:One is to choose appropriate mathematical methods to describe evidence, decisions, and supportConcepts such as degree;The second is to establish a fast, reliable, and easy to implement universal evidence combination algorithm structure.

The combination of evidence method has the following advantages over the embedding bundle method:

(1)For multiple typessensorThe physical relationship between data does not need to be accurately understood, that is, there is no need to accurately establish multiple sensor dataThe model of the body.

(2) Good passability enables the establishment of an evidence combination method that is independent of the background forms of various specific information fusion problems,Beneficial for designing universal information fusion software and hardware products.

(3)Artificial prior knowledge can be regarded as data information-Provide support for decision-making and participate in evidence combination operations.

threeArtificial Neural Network Method

The artificial neural network method imitates the structure and working principle of the human brain, designs and establishes corresponding machines and models, and completes themCertain intelligent tasks.

Neural networks can determine classification criteria based on the similarity of the samples received by the current system. This determination method mainly involves the following tableNowadays, in terms of network weight distribution, specific learning algorithms for neural networks can be used to acquire knowledge and obtain uncertainty inference mechanisms. Multiple neural networkssensorThe implementation of information fusion can be divided into three important steps:

(1)Select the topology structure of the neural network based on the requirements of the intelligent system and the form of sensor information fusion.

(2)The comprehensive processing of input information from each sensor is as follows:-A global input function and define its mapping as the mapping function of the relevant units,It reflects the statistical laws of the environment into the structure of the network itself through the interaction between neural networks and the environment.

(3)Learn and understand the sensor output information, determine the allocation of weights, complete knowledge acquisition and information fusion, and then interpret the input mode, converting the input data vector into high-level logic(Symbol)Concept.

Based on neural networkssensorInformation fusion has the following characteristics:

(1)Having a unified internal knowledge representation form, the network can acquire knowledge through learning algorithmssensorInformation is fused to obtain the parameters of the corresponding network, and knowledge rules can be converted into digital form to facilitate the establishment of a knowledge base.

(2)Utilizing external environmental information facilitates automatic knowledge acquisition and parallel associative reasoning.

(3)Being able to integrate complex relationships in uncertain environments into accurate signals that the system can understand through learning and reasoning.

(4)Due to the large-scale parallel processing capability of neural networks, the system's information processing speed is very fast.