The Improvement Of Passive Ranging Accuracy Tunning Based On Multi-Models

Infrared passive ranging system does not radiate energy outward,and only passively receives the infrared radiation from the target,then convert the received radiation into distance through some algorithm.Compared with radar,senor and other active ranging systems,infrared passive ranging system has the advantages of excellent concealment and strong anti-electromagnetic interference ability,so it has high military application value.Researchers have proposed a variety of infrared passive ranging methods,but most of them adopt the mechanism modeling method,and only use a single model.Although these methods have strong theoretical,but the calculation is more complex,and the application conditions are strict.Data-driven methods often use machine learning theory to establish input-output relationship with history data.In recent years,with the development of artificial intelligence and computer technology,a variety of efficient intelligent algorithms have emerged,which also provides a new way for the infrared passive ranging method.This paper mainly studies the modeling and optimization of infrared passive ranging system from the perspective of data-driven.Firstly,the characteristics of infrared radiation and its interaction with atmosphere are analyzed,and then we established atmospheric transmittance databases of 3.5?m ~4.0?m and 4.5?m ~4.7?m at different atmosphere,zenith angles and distances by MODTRAN.Secondly,the principle of infrared passive ranging system and it's parameter is introduced,Then,we deduced the relationship between the output voltage,distance and the radiation of the target,and the three-band output voltage data set is generated.Then,the characteristics of Data-Driven Modeling Method and ensemble learning theory is analyzed,and the GBDT,Xgboost,AdaBoost model and Elman neural network model are established with the dual-band voltage data set.Finally,we merge the established model in different methods.To further improve the accuracy of the model,this paper also proposes a classification-regression modeling method.Firstly,the training data are divided into different distance segment,then the classification model of the distance segment is established,and then the regression model for the specific distance segment is established according to the previous merging method,we also use the method to establish three band passive ranging model,the result shows that the classification-regression modeling method can improve the accuracy of the prediction.Finally,the cost time of different models is compared and analyzed.We also discussed the parameter extraction of gbdt and AdaBoost model and how to generate the two model in C language.