4.3. 1 Common image enhancement processing methods
In the application of remote sensing geology, image enhancement processing methods can be divided into two categories according to the main enhancement information content: wave (light) spectral feature enhancement and spatial feature enhancement.
4.3. 1. 1 image wave (light) spectrum feature enhancement processing
Image wave (light) spectrum feature enhancement processing is based on multi-band data, and the gray level of each pixel is transformed to achieve the purpose of image enhancement. The results of image enhancement are convenient to identify geological bodies, rock types, geological anomalies (such as alteration zones and thermal anomalies) and large-scale linear and annular structures with different properties.
(1) gray scale conversion method
When the histogram of the original image is narrow, the gray distribution is concentrated, and the image levels are few, the gray transformation is the most basic requirement. For images with gray levels close to normal distribution, linear stretching can usually achieve the purpose of improving the visual effect of images. For images with multi-modal histogram, some objects are too bright or too dark, so different gray transformation methods should be adopted according to the characteristics of the image, including piecewise linear stretching, histogram adjustment, Gaussian transformation and other nonlinear stretching.
The purpose of piecewise linear stretching is to effectively use the limited gray scale, divide the whole gray scale into several intervals, and linearly expand between the intervals to make the best use of the useful information in the image. Commonly used nonlinear transformations include exponential transformation (enhancing the high brightness value part of the original image), logarithmic transformation (enhancing the low brightness value part of the image), Gaussian transformation (enhancing the middle gray range of the image) and tangent transformation (enhancing the dark and bright areas of the image).
Histogram adjustment is to enhance an image by improving its histogram shape. Its principle is to use a transform function to act on the histogram of the original image to make it a histogram with a certain brightness distribution. This method focuses on expanding the interval between high-frequency brightness values, enhancing the contrast of ground objects contained in the middle of histogram, which is beneficial to the differentiation of geological bodies. The commonly used histogram adjustment methods are histogram equalization and histogram normalization.
(2) Ratio enhancement
Ratio enhancement is achieved by dividing the brightness values of pixels with the same name in different bands and generating a new ratio image. Ratio processing is particularly sensitive to geological information and has become one of the widely used methods in remote sensing geological image processing. Its basic functions are:
1) can amplify the spectral difference between rock and soil, which is beneficial to distinguish these features.
2) Eliminate or weaken the influence of environmental factors such as topography on similar lithology.
3) Extracting mineralization and alteration information.
4) The ratio color composite image can enhance the information of lithology and altered rocks.
(3) principal component transformation
Principal component transformation is a common method for multi-band remote sensing image enhancement. It is a multidimensional orthogonal linear transformation based on the statistical characteristics of images. The new component image after transformation reflects the total radiation difference and some spectral characteristics of ground objects, and also has the functions of separating information, reducing correlation and highlighting different ground objects. Using different new composition images for color synthesis can significantly improve the color enhancement effect and help to distinguish rocks. In practical application, the ratio or difference image and the original image are often used for principal component transformation, which will be beneficial to the extraction of some thematic information.
(4)IHS transformation
In colorimetry, converting RGB of a color image into brightness (I), chroma (H) and saturation (S) is called IHS transformation, while converting IHS into RGB is called inverse transformation. Using IHS transform and inverse transform, information fusion between multi-source remote sensing images, color enhancement of highly correlated image data, image feature enhancement and image spatial resolution improvement can be carried out. As shown in Figure 4. 1, it has played a certain role in strengthening the ring structure, rock mass and strata in the study area.
Fig. 4. 1 Contrast of image enhancement processing in Washixia region, Xinjiang.
4.3. 1.2 image space enhancement
Image spatial enhancement processing is to use the gray value of the pixel itself and its surrounding pixels to calculate and achieve the purpose of enhancing the whole image. The image enhancement results mainly highlight the spatial shape, edge, line and structural characteristics of geological bodies. Such as geological structure, linear body, landform and so on. Common image enhancement methods include data fusion and convolution enhancement.
(1) data fusion
There are mainly IHS fusion method, Brovey method and three-dimensional contrast enhancement fusion method. The key of IHS fusion method is to select the forward transformation and inverse transformation formulas according to the spectrum and ground coverage of the input image. Brovey method enhances image information by selecting the product of three normalized low-resolution band images and high-resolution images. The advantage of this method is that it can keep the original multi-spectral information while sharpening the image, and has a good enhancement effect on mountains, water bodies, vegetation and other ground objects. The fusion method based on three-dimensional contrast enhancement is to enlarge the gray difference of the same-name pixels in three low-resolution synthetic bands, and at the same time, it is required that the relative size relationship of the gray values of the same-name pixels in the three bands of the enhanced image remains unchanged, and the sum of the gray values of the three bands remains unchanged. For high-resolution images, gray linear stretching, texture energy enhancement and thinning are needed. This method enlarges the color difference between objects, reduces the correlation, and can separate the spatial information such as sharpness, gray level and color of the fused image and dynamically track it until satisfactory results are achieved. As shown in Figure 4.2, Figure 4.2(a) shows the original SPOT image, and Figure 4.2(b) shows the fusion result of SPOT data and TM multispectral data, which improves the reliability of remote sensing interpretation.
Figure 4.2 SPOT image fusion processing
(2) convolution enhancement
The boundary and linear features of ground objects usually show a certain spatial distribution frequency, which can be enhanced by spatial or frequency domain filtering. Convolution processing is a simple, effective and commonly used spatial filtering method. Convolution enhancement is a neighborhood processing technology, which is realized by convolution operation of the original image with a certain size template. By strengthening the boundary (or linear body) in different directions, the weight coefficient of each element in the template can be allocated according to a certain arrangement direction, and changing the template size and the difference of elements in the board can produce different effects. Generally, the larger the template, the greater the difference, the more obvious the enhancement of low-frequency rough structure characteristics, while the smaller the enhancement of high-frequency information (small faults, joints, cracks). Convolution enhancement is widely used in remote sensing geological image processing, because it has obvious effect on highlighting the boundary of geological bodies and linear fault structures or traces in a certain direction, and can also enhance some annular structures or traces.
The above processing methods are widely used in the extraction of lithologic strata and structural information. In practical application, images can be enhanced by various combinations and flexible means according to different geological and geographical conditions and image characteristics.
4.3.2 Feature information enhancement processing method
4.3.2. 1 lithologic information enhancement processing
The purpose of lithologic information enhancement processing is to extract rock types or type combinations through the selection of characteristic image processing methods. Its application principle is mainly based on the differences of different rocks in mineral composition, structural structure, rock surface structure, covering composition, water content and regional environment, and is realized by the laws of spectrum and texture information reflected by multi-band remote sensing data.
(1) band combination transformation method
A series of combinatorial algebraic operations are carried out on multi-band remote sensing images and single-band remote sensing images obtained by spatial registration of different sensors in the same area to enhance lithologic information. Take TM image data as an example:
1) The lithologic boundaries among granite belts, contact metamorphic belts and regional metamorphic rocks can be identified by using color synthetic images such as tm3,2, 1, tm4,3, 1, tm5,4,3, tm7,4, 1. As shown in Figure 4.3, the color composite images of TM5, 4 and 3 in Figure 4.3(a) obviously enhance the distribution characteristics of formation lithology in this area; The color composite images of TM7,4 and 1 in Figure 4.3(b) highlight the lithologic characteristics of dolomite.
2) Color synthesis in TM5/TM1,TM4/TM2 and TM5/TM7 bands can enhance and identify the mineralization information of carbonate rocks and clay. As shown in Figure 4.3(c), the geological boundary of dolomite is strengthened.
3) Using TM4×TM4/(TM4+TM5+TM7), TM4×TM5/(TM4+TM5+TM7) and TM4×TM7/(TM4+TM5+TM7) for color synthesis can enhance the change of light and shade between different lithology, reduce the difference of brightness values between yin and yang slopes caused by topographic relief, and improve the detail contrast of lithology. Both Figure 4.4(a) and Figure 4.4(b) enhance the distribution characteristic information of a certain lithologic profile.
4)TM5/TM 1, (TM5×TM7)/(TM 1×TM2), (TM7-TM 1)/(TM3+TM4) color composite images can enhance the change information of iron ions in Quaternary strata and surface, and highlight the structural structure of rocks. As shown in Figure 4.5, compared with Figure 4.5(a), Figure 4.5(b) has obvious application effect in extracting the change information of strata and rock mass in the area, distinguishing the main rock types and highlighting the ring structure.
(2) Spectral profiling method
When the rocks, strata and background in the study area are spectrally separable, that is, there is little isomorphism between them, lithologic thematic information can be extracted with the help of spectral profile knowledge. The main steps are as follows:
1) carries out spectral sampling on typical ground objects, such as exposed rocks, strata, snow, shadows, etc., extracts spectral profile curves, and finds spectral differences of different lithologic types.
2) Through the relationship between spectra, the characteristic lithology extraction models based on spectral knowledge are established respectively.
3) The exposed rock and stratum information are extracted according to the established model.
4) When there are many isomorphic phenomena between the spectra and backgrounds of different rocks and strata, it is necessary to use other knowledge of ground objects to extract them.
Fig. 4.3 Comparison of color composite images in different bands in Washixia, Xinjiang.
Figure 4.4 Comparison of Multi-band Correlation Ratio Enhancement Processing
Figure 4.5 Comparison of image enhancement processing
(3) Lithology identification based on texture of ground objects.
When the lithologic composition is complex and the distribution scale is larger than the spatial resolution of the sensor, the remote sensing image may record the structural composition information of the ground object, and its image has obvious texture characteristics. When there are texture features different from the background objects, lithologic information can be extracted by using the spectral features and texture features of the objects. The method of identifying lithology by texture is as follows.
1) Select a moving window with a certain size, calculate the texture characteristics of different features, and compare and analyze the texture characteristics of the rock type to be studied and the surrounding features. The main texture features are logarithmic variation function, average Euclidean distance method (first order), variance method (second order), slope (third order), kurtosis (fourth order) and * * * generating matrix method. The gray level * * * generating matrix can generate eight texture measures, which are local stationarity, contrast, dissimilarity, mean, standard deviation, entropy, angular second moment and correlation.
2) Analyze and study the texture index and image between the rock exposed area and the background object, find the correlation law between rock types and texture features, and extract rock information by using appropriate threshold recognition.
(4) Identifying lithologic information based on shape knowledge.
1) enhances the boundary between features and extracts the boundary information. Calculate the shape index. Mainly measure the index based on perimeter and area, the index based on area and the index based on area and area length.
2) According to the value of rock shape knowledge index, the lithology with different shape indexes is qualitatively identified and extracted, and some geological attribute information is given by combining the shape characteristics of different lithology.
(5) Principal component transformation multi-level information analysis identifies lithologic information.
Multi-level information decomposition technology based on principal component analysis is a common method to enhance weak geological and lithologic information. The implementation process of lithology enhancement and identification is as follows.
Statistical characteristics analysis of multi-band images. The statistical characteristics of multi-band image data are analyzed, and the gray dynamic range, mean and median of spectral images, correlation coefficient matrix and covariance matrix of band images are calculated.
2) The eigenvalues and eigenvectors of the covariance matrix of multi-band images are obtained, and the coefficient matrix A of KL transform is formed by the eigenvectors.
3) Post-processing of principal component transformation. According to the purpose of lithology identification and the analysis of the relationship between each principal component and matrix vector, the component image containing specific lithology information, the enhancement processing of the component image containing thematic information, the color synthesis processing of the component image and the information synthesis analysis of the component image and other processing results or band images are selected.
4) According to the analysis results of each principal component, the post-processing results of principal component images are compared with the unit results, and visual interpretation is carried out to determine the principal component images that can better reflect the rock information of the working area, and they are selected for color synthesis or information synthesis to enhance the weak information such as geology and lithology on the images.
(6) Enhance lithologic information by IHS transform method.
The new image generated by selecting appropriate algebraic operation for multi-band images can be transformed by IHS to highlight lithology. For example, the lithology and alteration characteristics related to mineralization in volcanic areas can be identified by using TM band ratio and IHS transformation.
The ratios of 1)TM5/TM7, TM3/TM4 and TM3/TM2 are assigned to red, green and blue respectively for IHS transformation.
2) On the transformed image, basalt with high Fe2O3 content has striking brown or red hue, and volcanic rocks with different lithology have different hues, which can be distinguished from each other; The distribution area of mineralized altered rocks containing clay minerals and trivalent iron oxides is characteristic yellow.
(7) carrying out optimal multi-level density segmentation on the remote sensing image to extract lithologic information.
The purpose is to extract and identify rock information by selecting the best remote sensing recognition image and the best multi-level density segmentation in arid areas with sparse vegetation and widely exposed bedrock.
1) uses Fisher criterion to segment the image density, and through histogram statistics, finds a segmentation method that minimizes the sum of intra-segment deviations and maximizes the sum of inter-segment deviations, which is called the optimal multi-level density segmentation method of the image.
2) Assign different colors to the segmented image according to the gray level from high to low, and determine the geological and lithologic attribute information of different colors according to the regional geological map.
(8) Automatic classification and identification of lithology
In arid and semi-arid areas, automatic lithology identification and mapping can be achieved by using spectral information of remote sensing images and unsupervised classification methods.
Taking TM or ETM+ data as an example, the main realization process of unsupervised classification method is explained:
1) Select three three-band combinations from TM or ETM+ multi-band images, so that the correlation between bands is small and the multiplexed bands are the least.
2) Enhance the contrast of images in all bands with balanced contrast enhancement technology, so as to optimize the contrast of each band and eliminate the possible color deviation in color synthesis.
3) RGB-IHS transform is used to generate chroma images for each three-band combination, and then the chroma images are synthesized separately to generate chroma composite images.
4) Using the interactive clustering technology of three-dimensional feature space, unsupervised clustering classification is carried out on chromaticity composite images.
5) The template histogram matching classification technology is used to classify the classified images in space, and the structure and mode of the interested categories are detected.
6) Smoothing and simplifying the classified images by using spatial filtering method and small class merging technology.
7) According to the spectral curve shape of field inspection and classification and referring to geological map, the classification is endowed with lithology or identified and described by other ground object types.
8) Interactive category editing. Different lithology representing different areas is decomposed by location by category area editing method, and categories with the same lithology or consistent feature type are grouped by category grouping method.
9) Use edge detection technology to detect the edge of ground objects.
10) carries out interactive coloring on the adjusted classified image, and overlays the gray image reflecting the terrain background on the lithologic classification map to form a lithologic image map.
(9) Rock type identification based on rock block classification
It is suitable for rock identification in arid and semi-arid bedrock exposed areas. Taking TM data as an example, the main implementation process is described in detail:
1) carries out terrain correction on TM images to generate digital apparent reflectance images R 1, R2, R3, R4, R5 and R7.
2) Using TM6 and R 1 ~ R7 to carry out unsupervised classification by spatial clustering method, and draw a plane classification map.
3) Using TM6 and R 1 ~ R7 data for supervised classification, firstly, using the known samples as the training area, and the samples in the training area are thick single rock blocks, simple lithologic combination rock blocks, complex lithologic combination rock blocks and symbolic thin rock blocks, and then extracting similar targets point by point to work out the scheme.
4) Statistic the average, minimum, maximum, standard deviation, covariance and other parameters of each category in the classified image.
5) Analyze and classify the textures, and compile the texture type plan.
6) Overlay the plans of unsupervised classification, supervised classification and texture classification, and compile the plan of rock remote sensing type through visual interpretation and fusion of human-computer interaction.
7) Rock mapping. Fill in the known rock attribute information in the blank area of the same kind, and fill in the unknown blank area after the field inspection determines the lithologic attribute.
(10) Lithology Identification of Hyperspectral Data
Using imaging spectral data, the spectral characteristics of rocks and single or multiple minerals are quantitatively detected, lithologic and mineral information is extracted and identified, and thematic lithologic and mineral maps are compiled. The main implementation method is:
1) to determine some iconic spectral characteristics of rocks and minerals in the work area.
2) Using hyperspectral imaging data to extract the spectral curves of ground objects, and comparing them with the field spectral curves of rocks and the typical curves measured by some marker minerals in the laboratory, the lithology and the existence of marker minerals can be determined semi-quantitatively.
3) Through the detection of lithology and marker minerals, the purpose of exploration and compilation of lithologic distribution map can be achieved.
Image enhancement processing of fault structure and geological boundary in 4.3.2.2
This paper mainly uses spatial filtering and automatic linear extraction to enhance or extract fault structure information.
(1) spatial direction filtering method
Directional filtering is carried out on the original image to highlight the texture information in a certain direction and enhance the spatial structure of geological bodies.
1) See Table 4. 1 Determine the filter operator according to the required direction information.
2) The multi-band image is transformed by principal component, and the edge gradient of first principal component image is enhanced by directional filtering.
3) Enhance the local edge gradient of the image and suppress the contrast of the whole image, and then combine some smoothing methods to enhance the structural alteration zone and ring structure.
4) Image contrast expansion. Stretching, histogram transformation, ratio, filtering, etc. It is used to highlight the linear, edge and texture features in the image and enhance the image features of lithology, linear structure and ring structure.
5) Qualcomm filter enhances the surface features with high spatial frequency, and extracts linear bodies (such as joints, cracks and fractures) from tens to hundreds of meters; Low-pass filtering enhances the surface features of low spatial frequency and extracts long and large-scale geological features such as fault zones and alteration zones.
6) Gaussian convolution filtering is used to highlight the details of the boundary contour of geological bodies and distinguish rocks with large texture differences.
(2) Fourier power spectrum texture enhancement method
1) Take a window image with a certain size and do the Fourier transform of rows and columns respectively.
2) Find the power spectrum matrix and perform logarithmic transformation.
3) Calculate the texture metric to form a texture image.
4) Texture image interpretation, extracting linear volume information and lithologic geological boundary.
(3) Statistical method of image texture
Through the change of structural characteristics, the difference of fault activity and the change of rock composition are inferred, the range of active fault zone is delineated and the fault activity mode is explained.
(4) Line-loop image feature method
1) performs Qualcomm filtering and linear image enhancement on the image.
2) Directional filtering is carried out in four directions: 22.5 ~ 67.5, 67.5 ~12.5, 292.5 ~ 337.5 and 337.5 ~ 22.5.
3) Calculate the linear image density and isodensity map of unit area (2.5km×2.5km).
4) Visually analyze the plan of linear and circular images, screen out non-geological edge points, overlap and merge them, and divide the area, band and grade of linear images, the spatial structure between circular images and their combination relationship.
5) Geological attribute interpretation of linear and circular images.
(5) Automatic extraction of linear bodies.
1) uses directional filtering method to enhance the edge gradient of the first component of KL transform of multi-band images.
2) Binarization the gradient image to extract the edge point image.
3) Human-computer interaction removes interference and isolated edge points.
4) Connecting and counting the linear bodies by using Hough transform, and outputting the distribution map and density map of the linear bodies.
5) Linear structure extraction and geological analysis.
(6) image brightness temperature method
Select the thermal infrared remote sensing images in appropriate seasons and time, and extract the structural information with the extreme value line of brightness temperature distribution of thermal infrared band images as the symbol.
(7) Multi-principal component analysis
Firstly, various methods are applied, including general principal component analysis, selected principal component analysis (characteristic principal component selection), band ratio and so on. , extract the weak geological structure information from the image as much as possible, and then extract the best or better thematic information for secondary processing. There are two processing methods: one is the combination or superposition of different colors to highlight the theme information; The second is to select the results that are most conducive to the extraction of thematic information and the original band for principal component analysis again, and to extract and enhance geological information for the second time.
(8) Structural information extraction method based on fusion processing.
Different sensors have different application characteristics due to different wavelength ranges, different geometric characteristics and different resolutions. Fusion processing based on different sensor images can integrate the advantages of different sensor images and improve the recognition ability of structural information. The following is an example of TM and SAR image fusion processing.
1) Firstly, filter the SAR image to eliminate noise.
2) Secondly, the single-band SAR image and the multi-spectral TM image are geometrically registered and fused, and the filtered SAR image is used to replace the I component for IHS transformation, and then the SAR image is transformed into the main component by TM3, TM4 and TM5. Finally, the G-component, TM4 band and the first principal component image transformed by IHS are color synthesized as geological interpretation images.
3) The fault structure information can be directly extracted from the fused image, and the hidden fault structure information can be extracted by using the certain penetrability of SAR images.
Comprehensive control of regional geological stability and auxiliary extraction of remote sensing information in 4.3.2.3
1) to obtain multi-temporal and multi-platform remote sensing satellite data and collect ground control point data and regional geological environment data.
2) carrying out geometric fine correction and registration on the image. Firstly, the topographic map is scanned with high precision to form a digital image; Then the digital topographic map is transformed by projection, registered and mosaic, and the regional image is synthesized and mosaic. Finally, DEM and three-dimensional topographic and geomorphological visualization images of geological active areas are established.
Conduct human-computer interaction interpretation. On the basis of finely calibrated digital satellite images, on the one hand, various image processing is carried out to enhance the information of tectonic active zones, landslides and their development environment; On the one hand, visual interpretation, determination of regional geological stability information, positioning on the computer, division of boundaries, making graphics. Obtain remote sensing interpretation information, combine with other environmental data, comprehensively process, analyze, compare and modify.
Extraction and Enhancement of Hidden Geological Information in 4.3.2.4
Using gravity and magnetic data and different types of remote sensing images to extract hidden geological information.
1) Determine the location (boundary) and depth of underground structure by using gravity and magnetic grid data and three-dimensional Euler deconvolution method.
2) Using remote sensing images to interpret the structural features of the earth's surface, superimposing the structural information of corresponding positions extracted from gravity and magnetic data on the remote sensing structural images, displaying the structures with different depths on the images respectively, and extracting the information of hidden geological bodies and structural zones with the help of the different depth information of the structures on the images.
4.3.3 Automatic extraction method of remote sensing geological information
The purpose of computer automatic information extraction is to quantitatively express the knowledge used by geological experts for visual interpretation, and fundamentally realize the automatic extraction of knowledge participation. The existing computer automatic information extraction methods mainly include spectral feature model method, computer automatic classification method and information extraction method based on spatial data mining and knowledge discovery.
4.3.3. 1 spectral characteristic model method
Generally, the remote sensing information model is established by statistical regression, and the model parameters are constantly adjusted according to the actual situation of the specific image, so that the model is finally suitable for the image. Remote sensing information model is a ground object inversion model extracted from existing ground experiments. Due to many factors affecting image data, the reflection of ground objects on satellite images is not one-to-one correspondence with the measured data on the ground. The effective combination of remote sensing information theory and actual map images can automatically extract thematic information with limited application scope and accuracy. Lithostratigraphic unit modeling technology is a spectral feature modeling method. The specific steps are as follows.
1) takes the multi-element black carbonaceous shale, ophiolite belt, migmatite belt, ultrabasic rock body and other mineral source beds, ore-bearing strata and rock strata with special image characteristics as basic units, and the gray value of multi-band remote sensing pixels is a function of bands, and different units have different function curves.
2) Statistic the spectral characteristics of units with certain geological significance to determine the brightness range of specific units in each band and the aggregation of the same unit category in multi-dimensional space.
3) According to the variation parameters (mean and standard deviation) of the unit category, the lithostratigraphic unit model based on the brightness value interval of remote sensing image is established, and the lithostratigraphic unit information is automatically extracted by using the input threshold parameters and multi-band remote sensing data.
4.3.3.2 classification
Classification method plays an important role in automatic extraction of remote sensing information. Its core is the automatic segmentation of remote sensing images. The existing computer automatic classification methods mainly use remote sensing image data, although sometimes other geoscience knowledge can be added automatically, but it is far from making full use of the knowledge applied by human brain in analyzing images, so it is difficult to achieve high accuracy. Automatic mapping lithology by classification method is the most complicated and difficult problem in remote sensing image processing. Automatic classification can play a good application role in extracting some specific target information distributed evenly in a large area, such as vegetation, water, land, ice and snow.
4.3.3.3 is based on data mining and knowledge discovery technology.
Based on data mining theory and knowledge discovery technology, the automatic extraction of remote sensing thematic information includes knowledge discovery, application of knowledge to establish an extraction model, and extraction of remote sensing thematic information using remote sensing data and models. In the aspect of knowledge discovery, it includes the spectral characteristics, spatial structure and morphology from a single remote sensing image, and the spatial relationship between objects. From multi-temporal remote sensing images, not only the above knowledge can be found, but also the knowledge of dynamic change process of ground objects can be further discovered. Discover all kinds of related knowledge from GIS database. Using the found partial knowledge, partial knowledge or all knowledge, the corresponding remote sensing thematic information extraction model is established, and the information can be automatically extracted from the application of single knowledge and single model to the comprehensive application of multi-knowledge and multi-model, and from the use of single data to the comprehensive use of multi-data.