| Evaluating the quality of on-board compression of domestic optical stereo mapping satellite remote sensing image and verifying whether the satellite image compression meets the requirements of mapping application are of great significance for elaborating the on-board compression criteria of mapping satellite and improving the application of domestic mapping satellite.To verify whether the quality of on-board JPEG-LS compression of optical stereo satellite remote sensing stereo image as well as multispectral images meet the requirements of 1:10 000 stereo mapping application,the paper aims at the GF-7 that will be launched in 2019,the evaluation of the effect of analog stereo images compression on the accuracy of basic image quality includes gray level,texture and correlation and the important mapping product of Digital Surface Model(DSM)generated by stereo images is tested.Aiming at analog multi-spectral remote sensing images,this paper analyses the image compression application ability from two aspects of image segmentation consistency and classification accuracy.In the aspect of image quality evaluation,besides extracting gray and texture features from spatial domain,this paper also proposes texture feature analysis before and after image compression based on frequency domain.In frequency domain,image texture characteristics can be better expressed by the commonly used spatial-domain-based evaluation index.In the aspect of DSM quality evaluation,DSM based on 5 meters and 2.5 meters grid resolution are extracted from stereo image pairs.A series of DSM generated from the reconstructed images with different compression ratios are compared with the DSM0 generated with the original images to get the difference images,and the accuracy of DSM products is evaluated according to the elevation error.Compared with DSM grid with the resolution of 10 m-25 m,DSM with 5 m resolution can more reflect the impact of data compression on the accuracy of DSM extraction accurately.In the aspect of multispectral image classification accuracy evaluation,this paper uses object-oriented classification method to carry out the impact analysis of supervisory classification of multispectral remote sensing images.To a certain extent,it can better solve the problems of "different objects with the same spectrum" and "same objects with the different spectrum" in high-resolution remote sensing images,and it can get more ideal classification results compared with the pixel-based classification method.Thus,it is convenient to further analyze the possible impact of on-board image compression on multi-spectral image applications.The experiment result shows that:(1)As to the JPEG-LS algorithm adopted in GF-7,when the compression ratio is less than 4:1,the image features and texture information are kept well,and in frequency domain,the roughness of image texture is easier to be recognized;The image correlation coefficient are greater than 0.997,the reconstructed image peak signal-to-noise ratio is greater than 50,and the image compression quality is good.(2)With the increase of compression ratio,the accuracy of DSM products is decreasing,when the compression ratio is 4:1,the root mean square error(RMSE)of DSM is 0.34 m,0.47 m and 0.82 m in urban and hilly,mountain and high mountain areas respectively,which meets the demand of DSM extraction-related mapping application in image compression;DSM with smaller grids not only expresses real terrain more precisely,but also reflects the influence of image compression on the accuracy of DSM extraction.(3)When the compression ratio is 3:1,the multispectral remote sensing image has a high segmentation consistency,the absolute change rate values of the total number of image segmentation objects and the average area of image segmentation objects are not more than 1%.The reconstructed multispectral image also has a high overall classification accuracy,and the Kappa coefficient is 0.91;When the compression ratio is 6:1,the multispectral remote sensing image has a low segmentation consistency.The Kappa coefficient after reconstructed image classification is 0.87.The suitable compression ratio for multispectral remote sensing images is 3:1.The result provides an important basis for the elaboration of satellite onboard compression and the analysis of application ability of image mapping. |
PDF Full Text Request