Remote sensing data preprocessing

Based on the remote sensing images of IKONOS before the March 16 earthquake in 2008 and QuickBird after the September 1 earthquake, this study extracted the remote sensing earthquake damage information based on multi-source and multi-temporal change detection technology.

The preprocessing of remote sensing earthquake damage information extraction data based on multi-source and multi-temporal change detection technology is different from the preprocessing of ordinary remote sensing image data, and it requires high quality of two-phase images (high accuracy in radiation correction, geometric correction and image registration) because it is directly related to the accuracy of subsequent earthquake damage information extraction. In order to meet the requirements of "rapidity, efficiency and accuracy", this section puts forward the image preprocessing technology flow of multi-source and multi-temporal remote sensing earthquake damage information extraction based on change detection technology (Figure 4-2), and achieved good results through experiments.

Geometric correction method of (1) linear equation without control points

The * * line equation is based on the strict mathematical transformation relationship between image coordinates and ground coordinates, and is a direct description of imaging space geometry. The correction process of this method needs digital elevation model, which can correct the projection difference and geometric deformation caused by terrain fluctuation to some extent. At present, all satellite remote sensing data are attached with the precise position, altitude, speed, solar altitude angle and attitude of satellite sensors, which are generally stored in the header file or RPC file of remote sensing images. It is very convenient to use the geometric correction method of * * line equation to carry out high-precision geometric correction and positioning without ground control points.

High-resolution remote sensing images have narrow width, high spatial resolution, relatively little influence by the tangent plane and curvature of the earth, and little geometric distortion inside the images. Therefore, generally, the geometric correction method of * * line equation can achieve good results after geometric correction of high-resolution remote sensing images. Because the geometric correction method of * * linear equation can correct the image only by providing the relevant parameters of satellite sensor flight, it saves the step of selecting control points, saves time and meets the requirements of "fast and accurate", so this study chooses the geometric correction method of * * * linear equation without control points for the first time to correct the image.

In ENVI software, read RPC file (. Txt format), and then geometrically correct the * * * line equations without control points in panchromatic band and multispectral band respectively in the geographic registration module.

Figure 4-2 Flow chart of image preprocessing technology for remote sensing earthquake damage information extraction

(2) Orthophoto correction

When surveying and mapping remote sensing images, due to the influence of various uncertain factors, such as the imaging mode of sensors, the change of external orientation elements, topographic relief, the curvature of the earth, atmospheric refraction and so on. The geometric shape of the image itself is often inconsistent with its corresponding feature shape, resulting in geometric deformation (distortion). The geometric deformation of remote sensing image refers to the deformation when the geometric position, shape, size, orientation and other characteristics of things in the original image are inconsistent with the expression requirements in the reference system. In order to eliminate the geometric distortion caused by these factors and pave the way for the subsequent image registration, it is necessary to use the DEM in the study area to carry out digital orthophoto correction on the images, and generate digital orthophoto images (DOM) before and after the earthquake respectively. The principle of digital orthorectification is the process of orthorectification by projecting an image through a digital element correction center (Chen Wenkai, 2007).

The orthorectification of this paper is completed in the orthorectification module of ENVI software. After obtaining the DOM before and after the earthquake, it is necessary to check its matching with DEM. The point-to-point error of feature points with the same name as DEM should not be greater than that specified in Table 4- 1. If it exceeds the requirements, it is necessary to carry out orthorectification again.

Table 4- 1 DOM and DEM point average error

(3) Image fusion

The DOM of panchromatic data and multispectral data is fused to form a fused image with high resolution spatial information and multispectral color information. Before fusion, multi-spectral data should be enhanced in color to expand the color contrast between different land types and highlight their color information. At the same time, the hue of the image is adjusted to improve the contrast and brightness of panchromatic data, enhance local contrast, highlight texture details and reduce noise. After fusion, it is necessary to check whether the image has ghosts and blurs, check the texture details and colors of the image, and judge whether the processing before fusion is correct. If there are the above problems, it is necessary to return to reintegration. If the brightness of the fused image is low and the gray level is narrow, linear stretching and brightness contrast can be used to adjust the tone, but attention should be paid to preserving the spectral information and spatial information of the fused data as much as possible.

In order to keep the multispectral features of the fused data (red, green, blue and near infrared bands) and to calculate the normalized vegetation index (NDVI) conveniently, a reduced resolution merging module is adopted under ERDAS software (this fusion method can keep the original multispectral features of the fused data). The panchromatic band and multispectral band of IKONOS and QuickBird data are fused respectively, and good results are obtained (Figure 4-3 and Figure 4-4).

Figure 4-3 IKONOS fusion image (1m)

Figure 4-4 quickbird fusion image (0. 6m)

(4) Image registration

DOM images before and after the earthquake have basically overlapped after geometric correction without control points and orthogonal projection correction of linear equations. Most of the ground objects can overlap well, but some targets are biased. In Figure 4-5, the image of QuickBird after the earthquake is on the left, and the image of IKONOS is on the right. In the area marked with black line in the middle, the overlapping effect of ponds is biased. In this case, it is necessary to register the images. Image registration, also known as accurate image correction, refers to the process of eliminating geometric deformation in images and producing new images that meet the requirements of some map projection or graphic expression.

Figure 4-5 Effect of DOM image superposition before and after the earthquake (black underlined area has deviation)

The image registration referred to in this section refers to the geometric registration of multiple images. Multi-source images refer to images in the same area at different times (multi-temporal images) or multi-source images obtained by different sensors. IKONOS and QuickBird images here belong to multi-source and multi-temporal remote sensing images. Geometric registration of multiple images refers to the exact coincidence of the same name image points of multiple images through geometric transformation, which is usually called relative registration. If multiple images obtained after relative registration are classified into the same map coordinate system, it is called absolute registration.

In this study, DOM images fused before and after the earthquake are registered in the image geometric correction module of ERDAS software. Taking the QuickBird image DOM after the earthquake as a reference, the quadratic polynomial correction model is selected to register the IKONOS image DOM before the earthquake. After manually selecting six control points with the same name to establish a polynomial model, ERDAS software will automatically find out the corresponding positions of control points in the image according to the model, and then only need to correct their positions in the image window, saving time. The residuals of registration control points with the same name shall meet the requirements in Table 4-2. * * * Select 20 ground control points (GCP), and the total root mean square error (RMSE) of these 20 control points is 1.0773. See table 4-3 for the coordinate values and RMSE of each ground control point. Finally, the nearest neighbor method is selected to resample the image.

Table 4-2 Residual of Registration Control Points

Table 4-3 Coordinates and RMSE of Ground Control Points

sequential

After geometric registration of DOM images, quality inspection and control are needed. First of all, the point-to-point error of feature points with the same name in DOM images before and after the earthquake should not be greater than that specified in Table 4-4. In addition, the matching between two DOM images and land use status map (LUDRG) should be checked, and the accuracy should not be greater than that specified in Table 4-5. If it does not meet the requirements, it is necessary to use the land use map to register the two DOM images for the second time.

Table 4-4 Registration Accuracy of Multi-temporal DOM Feature Points with the Same Name

Table 4-5 Accuracy of DOM relative to Land Use Status Map

(5) image radiation enhancement processing

Because the acquisition time of DOM images before and after the earthquake is different, the solar radiation received by the ground is different, and the pixels of high-resolution remote sensing images have strong spectral heterogeneity, so there must be some differences in appearance between the two DOM images, which has a negative impact on the detection of earthquake damage change information. In order to eliminate these adverse effects and improve the accuracy of earthquake damage information extraction, it is necessary to enhance the radiation of DOM images before and after the earthquake, mainly including adaptive filtering and histogram matching.

In order to control the random noise in high-resolution remote sensing data (random noise often affects the uniformity between classes and the stability of boundaries) and the strong non-uniformity of spectrum between pixels, it is necessary to filter the image spatially. In this study, Frost adaptive filtering in ENVI software is used to filter DOM images before and after the earthquake, which can reduce the spectral heterogeneity of pixels, smooth the image and keep the edges and textures of the ground class clear. Frost adaptive filtering is a kind of filter with weight as the adaptive adjustment parameter. It determines a weight for each pixel and then filters it one by one.

The histogram matching mentioned in this study refers to the mathematical transformation of the image lookup table, so that the histogram of all bands of a multi-spectral remote sensing image is similar to that of another remote sensing image. It is often used for stitching between adjacent images or preprocessing of dynamic change information detection of multi-temporal remote sensing images. Through histogram matching, the difference of spectral information between multi-source remote sensing images caused by solar altitude angle or atmospheric radiation can be eliminated (Dang Rongan et al., 2003).

Based on the DOM of QuickBird image after earthquake in ERDAS software, the histogram matching of each band of IKONOS image DOM before earthquake was carried out. From the fusion results in the previous section (Figure 4-3 and Figure 4-4), it can be found that there are thick clouds and shadows caused by clouds in the study area, which leads to the complete loss of information in the clouds and shadows and seriously affects the data quality. In this case, the method of removing thin clouds can't solve the problem, and the replacement method of removing thick clouds can't be used, because the subsequent work is to extract the change information of earthquake damage. If it is replaced by other image data (such as IKONOS data before the disaster), it will inevitably affect the accuracy of extracting the later change information. After comprehensive consideration, it was decided that the thick clouds and their shadow areas should be treated irretrievably, and the clouds and shadows were proposed separately through classification, and then a mask layer was established to exclude the above areas from the QuickBird image and not participate in the follow-up research. Before the earthquake, the same area was deleted from the image of IKONOS. See Figure 4-6 and Figure 4-7 for the effects of the first and second DOM images after radiation enhancement and removal of thick clouds and shadows.

Figure 4-6 IKONOS final DOM image (1m)

Figure 4-7 Final DOM image of QuickBird (0. 6m)