Comparison on linear feature real width and interpretation width using Landsat TM8 images and GF-1 images

Linear feature general exist in the nature and RS images as a type of land use. Linear feature's image feature is long and narrow object on the RS images, and it is visibility to the human eyes. On the other hand , a large number of linear feature hide in the mixed pixel in the RS images for their relative invisibility. In the surface feature interpretation, linear feature always includes to the surface feature to enlarge the achievement for the limited resolution. So, accurate linear feature interpretation can supply the surface feature result from deduction technology. Landsat TM8 image and GF-1 image have been extensively applied in different trade for their high quality and high resolution in several years, so as to agricultural RS field. In the crop area estimation, the accuracy of linear features extraction in Landsat TM8 image and GF-1 image can impact on the crop area estimation accuracy directly. So, the study of linear feature real width and interpretation width has a great significance for the crop area and yield estimation. GF-1 image has obvious advantage in linear feature interpretation for the higher resolution. Research areas were selected in Heilongjiang province, involving 56 counties and cities. The field investigation time was Sep 22-28, 2013, and 341 linear features were einvestigated. The a certain number of linear features in field investigation was selected in random, then recorded the width with the tape, at the same time ,use the Trimbes GPS positioning. Preliminary statistics the result of the Heilongjiang Province linear feature field investigation in 2013, then classification the linear feature with trend, type, real width. There are 3 catalogs and 13 type, including south-north trend, east-west trend, northeast-southwest trend, northwest-southeast trend, highway, field road, forest belt, ditch, 0~10m、10~20m、20~30m、30~40m、>40m and so on. 23 Landsat TM8 images and 14 GF-1 images were selected for the linear feature interpretation, imaging time concentrated in Jul 11,2013- Sep 18,2013. The primary compression package of the Landsat TM8 image's the fifth band TIFF file, the sixth band TIFF file, the forth band TIFF file were selected for layer stack, and then resampling the layer stack result to Albers conical projection, Krasovsky ellipsoid, Pulkovo 1942 coordinate system file. The .img file's named way is ___
.img. The .img file's band combination is R:1, G:2, B:3. The primary compression package of the GF-1 image's TIFF image was handled to receive the .img file with Albers conical projection, Krasovsky ellipsoid, Pulkovo 1942 coordinate system file. The .img file's named way is < satellite name and number >___< center point latitude>_< imaging time >_< projection >.img. The .img file's band combination is R:4, G:3, B:2. The study compared the 341 linear features' real width and interpretation width though statistics method in Heilongjiang province using 23 Landsat TM8 images and 14 GF-1 images. The results showed that all the trends linear feature's real width and interpretation width had a large standard deviation, and coefficient of variation of GF-1 images interpretation width's standard deviation and coefficient of variation were smaller than the Landsat TM8 images interpretation width's. According to the classification of linear feature type, highway and ditch's standard deviation and coefficient of variation had a vary widely to the field way and forest belt's, and GF-1 images interpretation width's standard deviation and coefficient of variation were smaller than the Landsat TM8 images interpretation width's. According to the classification of real width, because all the sample values are in the same interval, all the types had a small standard deviation and coefficient of variation except real width >40m, and GF-1 images interpretation width's standard deviation and coefficient of variation were smaller than the Landsat TM8 images interpretation width's. It means that RS satellite image's resolution determined the linear feature interpretation precision. 94 linear features had low interpretation precision with Landsat TM8 images interpretation (the absolute value of the exaggeration more than 50%), for 27.5660% of the entire linear feature, and in this part, interpretation width always larger than real width, and the category of 0-10 m had a worst interpretation precision, the category of trend and type had no influence to the linear feature interpretation accuracy. 29 linear features had low interpretation precision with GF-1 images interpretation, for 8.5044% of the entire linear feature, and in this part, interpretation width always much greater than real width.