Jitter Processing Of Remote Image Based On Convolution Neural Network

During the operation of remote sensor in orbit,the reaction flywheel,solar panel and other components themselves or in the process of orbital maneuver will produce vibration.Most of the vibration with larger amplitude and lower frequency can be measured by accelerometer,gyroscope and other attitude detection equipment,and then compensated and suppressed by the attitude correction mechanism;but the vibration with smaller amplitude and higher frequency will still be transmitted to the remote sensor through the platform,which makes the remote sensor vibrate weakly and weakens the stability of the remote sensor attitude.Due to the small field of view angle per pixel of high resolution remote sensor and the long distance from the object,the jitter will cause the blur of the scene image points in the exposure time and the degradation of the image quality.In this paper,convolution neural network is used to deal with the distortion of remote sensing image(1)In order to restore the distortion of RGB remote sensing image,firstly,according to the principle of RGB remote sensing image generation,the causes of distortion are analyzed,and the mathematical model of image jitter is established.Then the remote sensing image data set is preprocessed and enhanced.According to the mathematical model of image jitter,the three channels of RGB remote sensing image are separated and transformed into gray image.The training set and test set are established by applying distortion.By separating the three channels of RGB image,the confusion of the three channel features in the subsequent convolution process is avoided,and the multi-channel image processing is transformed into single channel image processing,which reduces the complexity of the processing.Then,in order to identify and detect the jitter value of image distortion,the forward path convolution network model and convolutional GAN model for single channel image processing are designed respectively,which can output the jitter vector and the repaired image.(2)By analyzing the jitter characteristics of satellite platform,the flutter information transmission between star sensor and satellite platform is realized;By extracting and transforming the flutter information of the blurred satellite image,the information that can improve the stability of the satellite platform and optimize the remote sensing image to the ground is obtained.Firstly,the flutter transmission relationship among satellite platform,remote sensing camera and star sensor is analyzed,and the feasibility of flutter information transmission among star sensor,satellite platform and earth remote sensing load is verified.Then,the center jitter trajectory of star sensor is extracted by convolutional GAN model,and the network function equivalent to point spread function is realized.Finally,the forward path convolution neural network model is used to get the pixel information of blurred image.By analyzing the flutter vector of remote sensing image distortion and transferring the flutter information between multi-sensor,the relevant vector information of micro flutter of satellite platform can be obtained,which not only provides a new solution for remote sensing image de-jitter processing,but also provides valuable information for improving the stability of satellite platform.