Research On Automatic Extraction Method Of Remote Sensing Linear Structure Based On Image And Band Optimizatio

Lineaments are a series of linear image features which are related to geological processes or controlled by geological structures on remote sensing images.It can be extracted by manual visual interpretation or computer automatic detection,but the latter is more efficient and can overcome the influence of human subjectivity.However,there were still the follow problems in the automatic extraction of remote sensing lineaments:(1)Most researches only pay close attention to the lineaments extraction result,but ignore to select the suitable optical satellite image(single phase,single band)for lineaments extraction through semi-quantitative evaluation;(2)the lineaments extracted were often not screened effectively and contain a large number of disorganized line segments,moreover,at the level of morphological characteristics,the connectivity of the lineaments was poor,then the correspondence between the lineaments and the geological structures was not well.The first problem is closely related to time phase factor and spectrum factor,while the second problem involves the scale dependence of remote sensing information extraction.The dissertation took Huize lead-zinc mine district in Yunnan Province as test area,and used Landsat-8(optical satellite image)and ASTER GDEM(digital elevation model)to carry out the research on the optical satellite image(single phase,single band)optimization and remote sensing lineament automatic extraction method.The main achievements were as follows:(1)Image optimization: firstly analyzed whether the solar illumination conditions during Landsat-8 in four time phases imaging could contribute to show structural landform outline based on the structural similarity,gray correlation degree and correlation between the illumination coefficient images(corresponding to Landsat-8 in spring,summer,autumn and winter)and the shaded relief image(enhancing main fault information),then analyzed the possible influence of surface vegetation coverage during Landsat-8 imaging in four time phases on geological structure information detection by normalized vegetation index and their corresponding spatial heterogeneity image.The research result showed that the difference in the morphological structure between the autumn illumination coefficient image and the shaded relief image was small,which was because the solar altitude was low during autumn Landsat-8 imaging,and the sunlight irradiation direction was roughly perpendicular to the direction of the main structure line in the test area,so the autumn Landsat-8 image would have strong stereo perception and show the structural landform outline clearly;in addition,the vegetation coverage on the surface in the test area was the most uniform in autumn,then the boundary lines that derived from the difference of vegetation coverage was the least obvious,so these natural linear elements that had nothing to do with geological structures would may cause least interference in the process of lineaments automatic extraction.Accordingly,the autumn Landsat-8 image was more suitable for lineaments extraction.(2)Band optimization: firstly analyzed the sensitivity of different bands to topographic relief based on the relationship between each band image(before and after topographic correction)and illumination coefficient image,then analyzed the corresponding relationship between the main linear traces of each band image after gray level differential and directional filtering and geological structure.The research result showed that NIR image(corresponding to autumn Landsat-8 OLI5)had the highest linear fitting accuracy and the strongest correlation with illumination coefficient image,and the correlation change after topographic correction was also the most prominent,so the pixel gray values of NIR image had very significant change tendency to topographic relief,in the area with prominent geological structure feature,which would make more obvious difference in stereo perception between the fault zone and the non-fault zone,then show the linear features that were related to the structural landform outline more effectively;in addition,the main linear traces of NIR image(corresponding to autumn Landsat-8OLI5)had the best correspondence with the main faults in the test area,and the artificial linear elements and natural linear elements that were unrelated to geological structure were not prominent,so the interference information was the least.Accordingly,autumn Landsat-8 OLI5 NIR image was more suitable for lineaments extraction.(3)Analyzed the influence of pixel scale on lineament extraction on single band image.After selecting the autumn Landsat-8 OLI5 NIR image,reduced the spatial resolution by resampling,then extracted the multi-scale lineaments in the test area through six conventional edge detection operators(four differential operators and two gray prediction models).The research result showed that,with the gradual reduction of the spatial resolution,a large number of microrelief traces were concealed,then macroscopic geological features tended to be relatively prominent,which made the line segments that were detected automatically by computer reduced gradually,some short line segments belonging to the same linear element and with collinearity characteristics connected gradually,and the connectivity of the remaining line segments enhanced.However,even if applying the Canny operator with superior performance,the extraction effect of lineaments from low-resolution Landsat-8 OLI5 NIR image after downsampling was still not ideal,moreover,due to the intense subjectivity of scale selection,it was difficult to acquire the optimal resolution for lineaments extraction.Therefore,it had great significance to realize adaptive spatial scale transformation in the process of edge detection for automatic extraction of remote sensing lineament.(4)Put forward a set of automatic extraction method for remote sensing lineament that could realize adaptive spatial scale transformation,which included two stage.(ⅰ)Stage one-multi-scale edge detection: after selecting the optical satellite image(single phase,single band)that were suitable for lineaments extraction,combine wavelet modulus maximum algorithm and two-dimensional maximum between-cluster variance algorithm for multi-scale edge detection,as long as the solar illumination conditions during imaging had little difference and all could effectively enhance the geological structure information,then the boundary lines(hue anomaly lines)that derived from the difference of vegetation coverage was not too prominent,even if the optical sensors were different,the lineaments that correspond to the main faults could be screened preliminarily from the NIR images and the images whose wavelength ranges are close to NIR spectrum of Landsat-8,ASTER,Sentinel-2,GF-1 without reducing the spatial resolution of the image,and their morphological characteristics were basically the same,but the higher the spatial resolution of the image,the more levels of wavelet transform had.(ⅱ)Stage two-terrain shadow aided tracking: firstly,acquire the shaded relief image that enhanced the geological structure information in main direction based on digital elevation model(DEM),then binarize the shaded relief image according to the gray value interval of the linear structural landform displayed by density segmentation,and generate the linear terrain shadows which were roughly parallel to the main structure line direction through a series of mathematical morphological filtering,subsequently superimpose the linear terrain shadows onto the main lineaments(need to filter out redundant short lines)extracted from the optical satellite image(single phase,single band)to further enhance the correspondence between the lineaments and the main faults.(5)Extracted four scales of lineaments in the test area from autumn Landsat-8 OLI5 NIR image by using the proposed method,then analyzed their geological characteristics.The research result showed that,in multi-scale edge detection of Stage one,when the level of wavelet transformation is low,the extracted line segments were mostly the fragmented large structural landform outlines and microrelief traces lines;with the increase of wavelet transformation level,the microrelief traces lines would be filtered out gradually,while the connectivity of main ridge lines and the large valley lines enhanced gradually,which would make the structural landform outlines clearer;when wavelet transformation reached Level 3,at the level of statistical characteristics,the correspondence between the lineament field and the geology field could be revealed by means of the two characteristic parameters of “NE-strike lineaments proportion” and “NE-strike lineaments fractal dimension ratio”;when wavelet transformation reached Level 4,at the level of morphological characteristics,there was a certain degree of correspondence between the lineaments that were screened preliminarily and the main faults in the test area,after terrain shadow aided tracking of Stage two,the correspondence would be further enhanced,then the geological significance of lineaments would be more explicit.The automatic extraction method for remote sensing lineament proposed in the dissertation had a certain universality,it could still achieve good effect in another two validation areas(Maoping lead-zinc mine district in Yunnan Province,Tayuan district in Heilongjiang Province)when selecting suitable image for the experiment.It should be pointed out that most of the extracted main lineament in the above three areas are the main outlines of linear structural landform(mountains and valleys)which have generative association with the geological structures,rather than the geological structures themselves,so there are bound to be some differences in morphological characteristics and spatial positioning between them;however,in the area with prominent geological structure feature,the form of structural landform and drainage pattern are often the outward manifestation of geological structural phenomena,which make the combinational relationship of mountains and valleys shown by these lineaments roughly reflect the regional geological structure framework.