Remote sensing ground object identification method based on DBN distribution integration and conflict evidence synthesis

Abstract

The invention discloses a remote sensing ground object identification method based on DBN distribution integration and conflict evidence synthesis, and belongs to the field of remote sensing ground object classification. The method comprises the steps that firstly, elevation features, spectral features and intensity features of LIDAR system data and multispectral images are extracted, and featurevectors are constructed; randomly selecting a certain number of selection features and sample experiment statistics to determine structure parameters of the distributed DBN network; constructing a plurality of independent and parallel DBN networks to obtain category probability as evidence information; and finally, calculating an evidence synthesis weight, and obtaining a decision-level fusion result according to a D-S synthesis principle. The advantages of distributed DBN integration and conflict evidence synthesis are fully utilized to mine and analyze airborne LIDAR data ground object information, and accurate classification of remote sensing data ground objects is realized. Experimental results show that the overall classification precision is obviously superior to that of classification results before sufficient synthesis.

Description

technical field [0001] The invention relates to remote sensing data object recognition technology, in particular to a remote sensing object recognition method based on DBN distribution integration and conflict evidence synthesis. Background technique [0002] The abundance of remote sensing data provides data resources for enhancing the accuracy of ground object classification and recognition, and can provide important decision support for industry applications, such as digital city construction, urban management, and natural disaster investigation. The LIDAR system can quickly and actively acquire the three-dimensional information of dense sampling points of ground objects, and obtain high-precision digital elevation models. At the same time, the multispectral images of the same scene acquired by the spectral camera have rich spectral information, which makes up for the single defect of obtaining ground object information. Using the information collected by the system as t...

AI Technical Summary

Problems solved by technology

However, the traditional evidence synthesis rules have essential flaws, which are caused by normalization, ignoring the contradictions and conflicts among evidences, and finally leading to serious paradoxes.

Many researchers have also proposed improved methods according to different application environments. Murphy proposed a method of modifying the model without changing the D-S composition rules, which averaged the basic trust distribution values ​​of the evidence, but did not consider the correlation of each evidence, making some deviations very large. Big data has a disruptive impact on fusion decision-making outcomes

For the LIDAR system, during the construction and training of the DBN network, the contribution degree of the randomly selected elevation features and spectral features is quite different, and the self-test results of the training network are quite different, and conflicts are inevitable. Therefore, there are big problems in the decision fusion results. There is an insurmountable bottleneck in the improvement of recognition accuracy

Images