Study On The Task Effectiveness Evaluation Method Of Agile Image Satellite Based On Machine Learning

With the development of satellite remote sensing technology,satellite remote sensing images are indispensable for both routine mission monitoring and spot detection.Agile image satellite has better mobility and quick response capability,which is an important direction of remote sensing satellite development.As more and more ordinary users can use satellite remote sensing systems,it takes longer time to obtain imaging results from observation requirements,long waits for task imaging results and performance results are unfriendly to the user experience,users want to get the performance result of the task before the task is planned,so as to have a expectation of the execution result of the task.Therefore,there is a growing need for users to predict the performance of submitted tasks.There is no literature on agile image satellite mission performance prediction and evaluation,it is of great significance to realize the pre-evaluation of the mission performance of the agile image satellite to improve the user experience of the satellite to earth remote sensing system and the development of the satellite to earth remote sensing system.In this paper,back-propagation neural network and support vector machine are used to regression models,by learning the historical task,performance value and obtaining the prediction model,the performance of the new task can be pre-evaluated.Firstly,this paper studies the content and data generation of mission performance data of agile image satellite,an agile image satellite mission performance data model composed of mission performance index and mission description model is established.Among them,the task completion rate,resolution and imaging duration were selected as the task performance indicators from three aspects of the task coverage efficiency,the task imaging quality efficiency and the task time efficiency;From the mission strip data and satellite orbit data,nine mission characteristics,such as the number of mission strip,strip length and orbit height,which have great influence on mission performance,were selected to form the mission description model;According to this data model,the task performance data of typical tasks are collected to complete the task performance data collection of agile image satellite,this data set is divided into training set and test set for the training and testing of machine learning algorithms.Secondly,research on agile image satellite tasks was conducted based on the assessment of task completion rate,the task completion rate was 0.5 as the threshold of task classification,and the tasks were divided into two types of tasks with a completion rate of [0.0,0.5)and[0.5,1.0].This paper designs the modeling process of classification model;Perform data statistical analysis,feature importance analysis and feature selection on task data;Then,the classification algorithms commonly used in machine learning were used to carry out classification experiments on the task data,and three algorithms with better classification effects,SVM,GBDT and Ada Boost were selected.The parameters of these three algorithms were adjusted and tested.The test results showed that GBDT algorithm had the best classification effect among the three algorithms;Finally,the GBDT algorithm is used as the agile image satellite task evaluation method based on task completion rate.Finally,this paper studies the pre-evaluation model of mission effectiveness of agile image satellite,and takes the average relative error of fitting results less than 10% and the maximum relative error less than 20% as the design requirements of the model,two regression models GBDT+BP neural network and GBDT+SVR were proposed and applied to the research of pre-evaluation model of mission performance.In this paper,the modeling process of regression model is designed;Perform data statistical analysis,data correlation analysis and feature selection for task data;Based on the BP neural network model and the SVR model,the regression fitting experiment was conducted on the three task performance indicators of task completion rate,resolution and imaging duration.The experimental results show that the two regression models cannot meet the design requirements.Then,combining the GBDT algorithm with these two models,two regression models GBDT+BP and GBDT+SVR are obtained.The experimental results show that the fitting results of the GBDT+BP neural network regression model meet the design requirements of the regression model in this paper,Finally,the GBDT+BP neural network regression model is used as the pre-evaluation model of the mission performance of the agile image satellite,and the research on the mission performance evaluation method of the agile image satellite based on machine learning is completed.