GISdevelopment.net ---> AARS ---> ACRS 1991 ---> Poster Session 2 The Study of the Correction Model of Daul - direction Scanning Error of TM A DATA Image Li Shu-kai, Liu Tong The Institute of Remote Sensing Application Chinese Academy of Sciences, Beijing, 100101 Abstract The sensors of TM image are different from MSS on LANDSAT. The MSS data is collected with mono-direction scanning and TM data with daual - direction scan. The sensor of TM is a improved sensor of MSS. Position deviation in scanning direction is produced by the difference of normal and reversal direction scanning of TM sensor. The deviation value is a complicate function of T ( data collection time of satellite ) in which there is some high - frequency elements. The charactertics of TM A DATA image is analysed first in this study. Using direct digital correlation method, the study of correction model of dual-direction scanning error of TM image has been finished primely. Some algorithms and software's are developed. The result is that 2.15 hours is necessary for the processing of correction of dual-direction scanning error with a scence TM A DATA. Preface The application of remote sensing image has been developed from the analysis of visual - interpretation and computer classification to a new period, in which compound analysis of multi-remote sensing image is developed with multi-criterion and multi-data resources. Positioning precession is distinctly more and more important. Factual positioning precision of rough correction image is unclear to say, where the relief displacement must be corrected in the complicated terrain area. In the edge of TM image , relief displacement of half pixel is produced by 117 meters height difference and 8.4 pixels by 2000 meters. These problems have to be considered when the compound analysis is done with the multi-remote sensing information and non-remote sensing information. Relief displacement can not be corrected with precise geometric processing of polynomial method. The best geometric precise processing model for relief displacement correction is the collincrity equations of central projection with parameters. The model of collinerity equations with parameters adapt to the original data which had never been corrected and projective relationship is central projection, that is, MSS and TM A DATA. The research of processing of MSS A DATA had already finished. The study of correction with dual-direction scanning error of TM A DATA is significant for precise correction of TM A DATA. The accurate geo-base for compound processing with multi-data resources also is the purpose and significance for the study. 1. The geometric feature analysis of LANDSAT TM A DATA The sensor of TM image on LANDSAT 4,5 are dual-direction line scanning sensors with a correction equipment of scan - Line behind the prime lens. The function is for forward movement compensation to the dual-direction scanning line. Then each scanning line is designed vertical to the satellite orbital trace ( figure 1). There are 16 scanning - lines for band 1 to band 5 and band 7,4 scanning - line for band 6. The data collection of TM is different from MSS. Image data of TM is collected by west to east scanning ( forward scan) and east to west scanning ( backward scan ) but MSS just with west east. Because of the difference of forward scan and backward scan, the object on ground is successive in the TM A DATA image. Fig 1 The correction of TM scan-line In figure 2, there are four scanning band ( 1,2,3,4) that each band is composed of sixty scanning line which is the data of forward or backward scan. A,B, and C is road, lake and river, Image of the road on the even band is not successive with the odd band. After verification, L1 is not equal to L2. The value of L is about 45±10 pixels with statistic result of 200 scanning bands. Fig 2 Un-successive ground object on TM A DATA 2. The principle and mathematic model of correction to the dual-dire The scanning rate of TM scan lens is about 13.9936 times per-second. The ground rate of LANDSAT 4,5 IS 6.7km per-second. Both of them is matched each other which means that the ground object between the scanning line ought to be successive. The reflecting-radiation energy distribution is complicated continuously in a band on a area of the ground. The shape and variation of grey level of the image are the important basis for distinction of the certain scale ground object. When the image of an object is un-successive, the method of visual reconstruction with TM A DATA is that moves the odd band line to the even band line till the object is successive. The function of manual visual interpretation is anologed by computer with the visual method in this study. In figure 3, the grey level curves show variation of the neighborhood line of neighborhood band to TM A DATA E.g., the sixteenth scanning line is marked L=16 and other so on. Line 16 and line 17 are a couple of neighborhood lines. The x coordinate is number of pixel on line. The y coordinate is the grey level of each image pixel. The grey level curve is with 1 to 260 pixels on band 4 of TM A DATA ( 123/31,1988). In figure 4, is very similar that the curve ABCDE on line 16 and curve A'B'C'D'E' on line 17. Similarity is also shown at the couple of line 32,33 and line 48, 49. The difference of x coordinate of A and A' on line 16,17 is about 45 pixels, the difference on line 32, 33 and line 48,49 are 48 and 560 pixels. Figure 3 Although ABCDE curve and A'B'C'D'E' curve are similar, but there is some difference be. tween them and that the similarity in a large extension is better than a small extension. The neighbor bands are almost scaned in the same time with a TM sensor on orbit. So the similarity of the curve reflect the similarity of the ground object . The direct digital correlation is the simplest digital method for similarity determination. The formulas are shown here: X and y is the average grey level of n pixels on line 16 and line 17 and so on. The variance and covariance of line 16 and line 17 are: The r is a correlation coefficient. The correlation relationship of scanning line can be expressed by coefficient r. The correlation coefficient r of limited length of two neighbor line is calculated to compare. Then the position with the mumxam value of r is where we seek. IN figure 3, the similarity of the grey level curve with 100 pixels can be indicated enough about it. Of course, that of 200 pixels maybe better than 100 pixels, but more is CPU time. The CPU time and correlation precision are connected with how many pixels in a grey level curve. In this experiment the length of grey level curve is with 100 pixels. The number of correlation coefficient r is determined with the searching extent of max correlation coefficient r. The statistic deviation with 200 scanning bands is about 45± 10 pixels, which means that searching extent is 20 pixels. The true correlation position is where the correlation coefficient r is max by which the value of movement between odd band and even band is determined. 3. Algorithms and experiments The principle and mathematic model of dual - direction scanning error of TM A DATA showed before. The ground object is surecessive in a band of TM A DATA . The scanning lines which correlation coefficient r is calculated are No 16 of odd band and No 1 of even. The pixel sequence for correlation is selected in the middle and two sides of band with 160 pixels. There are 358 bands in a scence TM A DATA. 63 times is calculated to 100 pixels for a pair scanning line. The correlation calculation of three place in a pair line may be improved the correlation precision in which the movement value of max r is the value for the odd band to even band, There are some other normal algorithers for precision improvement which will not be introduced here. The TM A DATA of this experiment was obtained in 1988 around Beijing. The orbit number is 123 / 31. There are 6656 pixels in each line. There are 5728 scanning lines in a scence TM image. The CPU time is about 2.15 hours with computer VAX - 11-785. the photos comparing shown in picture 1 and 2. 4. Conclusion and development in future It is suitable that the correction model of direct digital correlation for dual - direction scanning error of TM A DATA. The principle is the bionics of man - kind eyes. The direction digital correlation is the simplest method and efficient mathematic methods. The prime result had been achieved with in 2.15 hours on the VAX - 11-785 computer to a sence TM A DATA. Which the un-successive objective can not be fount with eyes. In this study the un-successive problem of object is solved to the TM dual-direction scanning error which did not be corrected in the primely data. The concept of this problem is different from geometric processing with / not with the altitude data of sensors. In the area with small grey contrast of ground feature, when the correlation coefficient r was calculated, there are two processing problems. One is how to improve correlation precision. Another is how to select the strong correlation extensive automatically. The result can be adapted in every kinds line scanning image (include CCD array image), and also include the DATA processing of China Earth Resource Satellite ZY-1. Original image of TM A Data Image corrected by correction model of daul-direction scanning error References WangZhizuo "The principle of photogrammetry" 1979.