The Research On Space Target Fusion Recognition And Knowledge Base Design

In recent years,as humans have explored outer space more frequently,the types and numbers of space targets have risen sharply.The need for space target recognition has also emerged.The space target recognition technology obtains the echo data of the space target,extracts the characteristics that reflect the characteristics of the space target,and confirms the target category through the classifier.Target fusion recognition technology combines target recognition methods with information fusion theory,and improves the accuracy and reliability of spatial target recognition by making full use of the complementary information between fusion information.This thesis focuses on the problem of spatial target recognition,and studies the method of spatial target feature-level and decision-level fusion recognition.At the same time,considering the requirements of target data retrieval and storage in spatial target recognition,this thesis uses Qt Creator development software to design a knowledge base for spatial target information.The main content of the thesis is as follows:1.Aiming at the task of spatial target feature fusion recognition,this thesis proposes a deep feature fusion recognition method based on adaptive weighted summation.In the process of feature fusion recognition,different features have different contributions to the recognition task.Traditional feature fusion methods take less consideration of such differences between features,resulting in limited recognition performance of fusion features.This thesis proposes a method to measure the contribution of the feature by introducing the attention module in the feature fusion process,and use the weighted sum method to obtain the fusion feature,and finally realize the joint optimization of the fusion network parameters based on the unified cost function.Experimental results based on electromagnetic simulation data of space targets show that the recognition rate of the feature fusion method proposed in this thesis is 94.16%,which is improved to a certain extent compared with other feature fusion methods.2.In order to further improve the recognition performance,this thesis combines the echo characteristics of the radar wide and narrow alternating working mode,and proposes a comprehensive recognition scheme based on the high and low resolution echo of space targets.This scheme performs feature-level fusion judgment on the features extracted from high-resolution and low-resolution echoes respectively;then performs decision-level fusion on the decision vectors of high-resolution and low-resolution echoes to obtain the final recognition result of the space target.Because the radar wide and narrow alternate working mode can obtain the wide-band radar echo and the narrow-band radar echo of the space target at intervals,and the wide-band and narrow-band radar echo contains the identity information of the target under different radar resolutions and attitudes.Decision-level fusion of the decision result of narrow-band radar echo and wide-band radar echo can make full use of the information related to the target category in the wide-band and narrow-band radar echo.In the decision-level fusion part,this thesis implements three decision-level fusion methods as follow: decision template method,D-S evidence theory and random forest,and verifies the performance of each method.Experimental results based on electromagnetic simulation data of space targets show that the proposed scheme can effectively improve the performance of space target recognition.Among them,the comprehensive recognition scheme based on the random forest decision-level fusion method has an average recognition rate of 97.08%,The feature-level fusion recognition performance of echo data is improved by about 3% compared to only using wide-band.3.Research the software design of the space target knowledge base.The acquisition and storage of spatial target knowledge is very important during the task of spatial target fusion recognition.In order to facilitate the management and storage of spatial target knowledge,this thesis uses Qt Creator to design the spatial target knowledge base based on the Visual Studio development platform.The completion of the design of the knowledge base mainly went through the steps of spatial target information collection,data structure design,development platform construction,database implementation and data visualization.The knowledge base of space targets includes 22 types of typical high and low orbit space targets at home and abroad,covering the structural characteristics of space targets,orbit parameters,simulation data,measured data,picture information and other data,in addition,the knowledge base software also supports data updates,retrieval,maintenance,visualization and other functions.