基于ENVI的极化SAR数据处理流程介绍
ENVI Tutorial: Polarimetric SAR Processing and
Analysis
Table of Contents
O VERVIEW OF T HIS T UTORIAL (2)
Background: SIR-C/SAR (2)
P REPARE SIR-C D ATA (3)
Optional: Read a SIR-C CEOS Data Tape (3)
Optional: Multilook SIR-C Data (3)
S YNTHESIZE I MAGES (4)
Default Polarization Combinations (4)
Other Polarization Combinations (4)
Display Images (5)
Define ROIs for Polarization Signatures (6)
Extract Polarization Signatures (6)
Adaptive Filters (8)
Slant-to-Ground Range Transformation (9)
Preview CEOS Header (9)
Resample Image (9)
Texture Analysis (10)
Create Color-coded Texture Map (10)
Image-Map Output (11)
Overview of This Tutorial
This tutorial demonstrates the use of ENVI’s tools for analyzing polarimetric synthetic aperture radar (SAR) data. You will multilook Spaceborne Imaging Radar-C (SIR-C) from Death Valley, California; synthesize images, define ROIs for (and extract) polarization signatures, use adaptive filters, perform slant-to-ground range transformation, use texture analysis, and create an output image-map.
Files Used in This Tutorial
ENVI Tutorial Data DVD: envidata\ndv_sirc
File Description
ndv_l.cdp L-band SIR-C subset in ENVI compressed data product (.cdp) format
pol_sig.roi Region of interest (ROI) file
Background: SIR-C/SAR
SIR-C is a polarimetric SAR instrument that uses two microwave wavelengths: L-band (24 cm) and C-band (6 cm). The SIR-C radar system was flown as a science experiment on the Space Shuttle Endeavor in April (SRL-1) and October 1994 (SRL-2), collecting high-quality SAR data over many sites around the world. (A second radar system, XSAR, was also flown on this mission, but these data are neither discussed nor processed here.) Additional information about SIR-C is available on the NASA/JPL Imaging Radar Home Page at https://www.wendangku.net/doc/5318552743.html,/.
Prepare SIR-C Data
The data used in this tutorial are a subset of L-band Single Look Complex (SLC) SIR-C data that cover the northern part of Death Valley, including Stovepipe Wells, a site of active sand dunes and extensive alluvial fans at the base of mountains. These data were preprocessed by reading and subsetting from tape and multilooking (averaging) to 13 m square pixels. The data are provided in ENVI compressed data product (.cdp) format. This non-image format is similar to the tape format and cannot be viewed until images are synthesized for specific polarizations.
The first two functions described in this example—reading the data tape and multilooking—were already applied to the SIR-C data. The steps are provided here only for completeness if you want to learn more about the processes. Skip to the Synthesizing Images section if you are not interested in data preparation.
Optional: Read a SIR-C CEOS Data Tape
1.From the ENVI main menu bar, select File→Tape Utilities→Read Known Tape Formats→SIR-C CEOS.
The SIRC Format - Load Tape dialog appears.
2.Enter the tape device name and accept the default record size of 65,536. Click OK. The tape is scanned to
determine what SIR-C files it contains. A dialog appears to let you select the desired datasets. By default, ENVI reads all of the data files on the tape.
3.If you do not want to read all of the data files, click Clear, then select the check box next to each desired file.
Click OK.
4.You can independently subset and multilook the selected data files as they are being read from tape. However,
you should perform multilooking on disk (unless you have insufficient disk space) as this function is extremely slow from tape.
5.Select a filename, then click Spatial Subset or Multi-Look to enter parameters for the data file. Enter an
output filename. Each input file must have an output filename. By convention, the output filenames should take the form filename_c.cdp and filename_l.cdp for the C- and L-bands, respectively. The SIR-C data are
read from the tape, and one compressed scattering matrix output file is created for selected each dataset.
Optional: Multilook SIR-C Data
Multilooking is a method for reducing speckle noise in SAR data and for changing the size of a SAR file. You can multilook SIR-C data to a specified number of looks, number of lines and samples, or azimuth and range resolutions. The SIR-C file used in this tutorial was a single-look dataset with a range resolution of 13 m and an azimuth size of 5 m. Multilooking has already been performed in the azimuth direction to make 13 m square pixel sizes. Instructions on multilooking are included here only for completeness.
1.From the ENVI main menu bar, select Radar→Polarimetric Tools→Multilook Compressed Data→SIR-
C Multilook. An Input Data Product Files dialog appears.
2.Click Open File and select an input file. ENVI detects whether the file contains L- or C- band data and displays
the filename in the appropriate field of the dialog. Click OK.
3.Select the file to multilook by selecting the check box next to the name. You can select multiple files.
4.Enter any one of three values—number of looks, number of pixels, or pixel size—and the other two are calculated
automatically. Integer and floating-point number of looks are supported.
5.Enter the desired Samples (range) and Lines (azimuth) values.
6.Enter a base filename and click OK.
Synthesize Images
The SIR-C quad-polarization data provided with this tutorial and available on tape from JPL are in a non-image, compressed format. Accordingly, images of the SIR-C data must be mathematically synthesized from the compressed scattering matrix data. You can synthesize images with any transmit and receive polarization combinations you want.
1.From the ENVI main menu bar, select Radar→Polarimetric Tools→Synthesize SIR-C Data. An Input
Product Data Files dialog appears.
2.Click Open File. A file selection dialog appears.
3.Navigate to envidata\ndv_sirc and select ndv_l.cdp. Click Open. When the filename appears in the
Selected Files L: field, click OK. The Synthesize Parameters dialog appears.
Default Polarization Combinations
Four standard transmit/receive polarization combinations—HH, VV, HV, and TP—are listed in the Select Bands to Synthesize list of the Synthesize Parameters dialog. By default, all of these bands are selected to be synthesized.
1.Enter ndv_l.syn in the Enter Output Filename field.
2.Click the Output Data Type drop-down list and select Byte. This scales the output data to byte values. (If you
will be performing quantitative analysis, the output should remain in floating-point format.) Click OK. After
processing is complete, four bands corresponding to the four polarization combinations are added to the Available Bands List.
Other Polarization Combinations
The transmit and receive ellipticity and orientation angles determine the polarization of the radar wave used to synthesize an image. The ellipticity angle falls between -45 and 45 degrees and determines the “fatness” of the ellipse. The orientation angle is measured with respect to horizontal and ranges from 0 to 180 degrees. You can synthesize images of non-default polarization combinations by entering the desired parameters as follows.
1.From the ENVI main menu bar, select Radar→Polarimetric Tools→Synthesize SIR-C Data. The file
ndv_l.cdp should still appear in the Selected Files field. Click OK. The Synthesize Parameters dialog appears.
2.Enter -45 in both the Transmit Ellip and Receive Ellip fields and 135 in the Transmit Orien and Receive
Orien fields.
3.Click Add Combination. This will produce a right-hand circular polarization image.
4.Enter 0 in both the Transmit Ellip and Receive Ellip fields and 30 in the Transmit Orien and Receive Orien
fields.
5.Click Add Combination. This will produce a linear polarization with an orientation angle of 30 degrees.
6.Click Clear under the list of polarization combinations to turn off synthesis of the standard polarization bands,
which have already been generated.
7.Select the Yes radio button for Output in dB? This will produce images that are in decibels with values typically
between –50 and 0.
8.In the Enter Output Filename field, enter ndv_l2.syn and click OK. After processing is complete, two bands
corresponding to the polarization combinations are added to the Available Bands List.
Display Images
1.In the Available Bands List, select [L-TP] under ndv_l.syn and click Load Band. The SIR-C, L-band, total-
power image appears in a new display group.
2.From the Display group menu bar, select Enhance→Interactive Stretching. A histogram plot window
appears, which shows the current stretch (between the vertical dotted lines on the input histogram) and the
corresponding DN values in the text fields.
3.Drag the dotted vertical lines to change the stretch, or enter the desired DN values into the appropriate fields.
4.Enter 5 in the left Stretch field and 95 in the right field.
5.From the histogram menu bar, select Stretch Type→Gaussian. Click Apply. A Gaussian stretch is applied
with a 5% low and high cutoff.
6.Generate and compare linear and square-root stretches.
7.To display a color composite, select the RGB Color radio button in the Available Bands List. Select
[L-HH], [L-VV], and [L-HV] in sequential order under ndv_l.syn.
8.Click Display #1 and select New Display. Click Load RGB to display the HH band in red, VV in green, and HV
in blue. The color variations in the images are caused by variations in the radar reflectivity of the surfaces. The bright areas in the sand dunes are caused by scattering of the radar waves by vegetation (mesquite bushes). The alluvial fans show variations in surface texture due to age and composition of the rock materials.
9.Adjust the stretch as desired (Gaussian and square-root stretches work well on all three bands).
10.Close the histogram plot window and Display #2 when you are finished. Keep Display #1 open for later exercises.
Define ROIs for Polarization Signatures
You can extract polarization signatures from a SIR-C compressed scattering matrix for a region of interest (ROI) or a single pixel in a polarimetric radar image. Define ROIs by selecting pixels or by drawing lines or polygons within an image.
1.From the Display group menu bar, select Overlay→Region of Interest. An ROI Tool dialog appears.
2.Four ROIs were previously defined and saved for use in extracting polarization signatures for this tutorial. From
the ROI Tool dialog menu bar, select File→Restore ROIs. A file selection dialog appears.
3.Select pol_sig.roi. A dialog box appears, stating that the regions were restored. Click OK.
4.Regions named veg, fan, sand, and desert pvt appear in the table in the ROI Tool and are drawn in the display
group.
5.To draw your own ROI, select ROI_Type→Polygon, Polyline, or Point from the ROI Tool menu bar.
6.Click New Region, enter a name, and choose a color.
Draw polygons by clicking the left mouse button in the display group to select the endpoints of line
segments, or by holding down the left mouse button and moving the cursor for continuous drawing. Click
the right mouse button once to close the polygon and a second time to accept the polygon.
Draw polylines in the same manner as polygons. Click the left mouse button to define the line endpoints
and click the right button to end the polyline and a second time to accept the polyline.
Point mode is used to select individual pixels. Click the left mouse button to add the pixel currently under
the cursor to the ROI.
You can select multiple polygons, lines, and pixels for each ROI.
7.Repeat Step 6 to draw a second ROI. You can save the ROIs to a file and restore them later by selecting File →
Save ROI from the ROI Tool dialog menu bar.
Extract Polarization Signatures
Polarization signatures are 3D representations of the complete radar scattering characteristics of the surface for a pixel or average of pixels. They show the backscatter response at all combinations of transmit and receive polarizations and are represented as either co-polarized or cross-polarized. Co-polarized signatures have the same transmit and receive polarizations. Cross-polarized signatures have orthogonal transmit and receive polarizations. Polarization signatures are extracted from the compressed scattering matrix data using the ROIs for pixel locations. Polarization signatures are displayed in viewer dialogs, as shown on the next page. To extract your own polarization signatures, perform the following steps.
1.From the ENVI main menu bar, select Radar→Polarimetric Tools→Extract Polarization Signatures→
SIR-C. The filename ndv_l.cdp should appear in the Input Data Product Files dialog. If not, click Open File and select this file. Click OK. The Polsig Parameters dialog appears.
2.Select the four ROIs (veg, fan, sand, and desert pvt) by clicking Select All Items.
3.Select the Memory radio button and click OK. Four Polarization Signature Viewer dialogs appear, one for each
ROI. The polarization signatures are displayed as 3D wire mesh surface plots and as 2D gray scale images. The X and Y axes represent ellipticity and orientation angles, respectively. You can selectively plot the vertical axis as intensity, normalized intensity, or dB by selecting Polsig_Data from the Polarization Signature Viewer dialog menu bar.
4.Polarization signature statistics appear at the bottom of each Polarization Signature Viewer dialog. Notice the
range of intensity values for the different surfaces. The smoother surfaces (sand and desert pvt) have low Z values. The rough surfaces (fan and veg) have higher Z values. The minimum intensity indicates the pedestal height of the polarization signature. The rougher surfaces have more multiple scattering and therefore higher pedestal heights than the smoother surfaces. The shape of the signature also indicates the scattering
characteristics. Signatures with a peak in the middle show a Bragg-type (resonance) scattering mechanism.
5.In any given Polarization Signature Viewer dialog, change the Z-axis by selecting Polsig_Data→Normalized
from the Polarization Signature Viewer dialog menu bar. This normalizes the signature by dividing by its
maximum; the signature is plotted between 0 and 1. This representation shows the difference in pedestal heights and shapes better, but it removes the absolute intensity differences.
Alternately, select Polsig_Data → Co-Pol and Cross-Pol to toggle between co-polarized and cross-
polarized signatures.
https://www.wendangku.net/doc/5318552743.html,e the left mouse button to drag a 2D cursor on the polarization signature image on the right side of the plot.
Note the corresponding 3D cursor in the polarization plot.
7.Click-and-drag any axis to rotate the polarization signature.
8.You can optionally output the signatures to a file or printer by selecting File→Save Plot As or File→Print
from the Polarization Signature Viewer dialog menu bar.
9.Close the Polarization Signature Viewer and ROI Tool dialogs when you are finished.
Adaptive Filters
Adaptive filters are used to reduce the speckle noise in a radar image while preserving the texture information. Statistics are calculated for each kernel and used as input into the filter, allowing the filter to adapt to different textures within the image.
1.From the ENVI main menu bar, select Radar→Adaptive Filters→Gamma. A Gamma Filter Input File dialog
appears with a list of open files. You can apply a filter to an entire file or to an individual band.
2.In the Gamma Filter Input File dialog, click the Select by toggle button to choose Band.
3.Select [L-HH] under ndv_l.syn and click OK. The Gamma Filter Parameters dialog appears.
4.Accept the default values, and select the Memory radio button. Click OK.
5.In the Available Bands List, click Display #1 and select New Display. Select the Gray Scale radio button,
select the new band name (Gamma), and click Load Band.
6.From the Display group menu bar, select Enhance→[Image] Square Root.
7.In the Available Bands List, click Display #2 and select Display #1. Select [L-HH] under ndv_l.syn, and
click Load Band.
8.From the Display #1 menu bar, select Enhance→[Image] Square Root.
9.From any Display group menu bar, select Tools→Link→ Link Displays. The Link Displays dialog appears.
Click OK to link the gamma-filtered L-HH image (Display #2) with the original L-HH image (Display #1).
10.Click in an Image window to toggle between the two images, using the dynamic overlay feature. The figure below
shows a portion of the original image (left) and the gamma-filtered image (right).
11.Close Display #2 when you are finished. Leave Display #1 (ndv_l.syn) open for the next exercise.
Slant-to-Ground Range Transformation
A radar system looks to the side and records the locations of objects using the distance from the sensor to the object along the line of sight, rather than along the surface. An image collected using this geometry is referred to as a slant range image. Slant range radar data have a systematic geometric distortion in the range direction. The true, or ground range, pixel sizes vary across the range direction because of the changing incident angles. This makes the image appear compressed in the near range, relative to what it would look like if all of the pixels covered the same area on the ground. Slant-to-ground range correction for SIR-C is performed on synthesized images. In other words, the correction is not performed on the entire SIR-C compressed data product file. However, this file does store the required information in the CEOS header about the sensor orientation.
Preview CEOS Header
1.From the ENVI main menu bar, select Radar→Open/Prepare Radar File→View Generic CEOS Header. A
file selection dialog appears. You must select the original unsynthesized data file from which to extract the
necessary information.
2.Select ndv_l.cdp and click Open. A CEOS Header Report dialog appears. Scroll down and note that the line
spacing (azimuth direction) is 5.2 m, while the pixel spacing (slant range direction) is 13.32 m. Close the CEOS Header Report dialog when you are finished reviewing it.
Next, you will use the Slant-to-Ground-Range function to resample the image to square 13.32 m pixels, thus
removing slant range geometric distortion.
Resample Image
3.From the ENVI main menu bar, select Radar→Slant to Ground Range→SIR-C. A file selection dialog
appears.
4.Select ndv_l.cdp and click Open. The Slant Range Correction Input File dialog appears.
5.Select ndv_l.syn and click OK. The Slant to Ground Range Correction Dialog appears. ENVI automatically
populates the Instrument height (km), Near range distance (km), and Slant range pixel size (m) fields with
information from the CEOS header.
6.Enter 13.32 in the Output pixel size (m) field to generate square ground-range pixels.
7.From the Resampling Method drop-down list, select Bilinear.
8.In the Enter Output Filename field, enter ndv_gr.img. Click OK. The input image is resampled to square
13.32 m pixels. Four new bands appear in the Available Bands List. Band 1 of the resampled image corresponds
to the L-HH band of the original, slant-range image (ndv_l.syn), Band 2 corresponds to L-VV, etc.
9.In the Available Bands List, click Display #1 and select New Display.
10.Select a band from the resampled image and click Load Band. The resampled image appears in Display #2.
Make sure Display #1 (ndv_l.syn) shows the corresponding polarization band.
https://www.wendangku.net/doc/5318552743.html,pare the two images.
12.When you are finished comparing images, close Display #2. Keep Display #1 (ndv_l.syn) open for the next
exercise.
Texture Analysis
Texture is a measure of the spatial variation in the gray levels in the image, as a function of scale. ENVI calculates texture based on a processing window size you specify. The texture measures demonstrated in this tutorial are Occurrence Measures, including data range, mean, variance, entropy, and skewness. These terms are explained in ENVI Help. Texture is best calculated for radar data with no resampling or filtering applied.
1.From the ENVI main menu bar, select Radar→Texture Filters→Occurrence Measures. A Texture Input
File dialog appears.
2.Click the Select By toggle button to choose Band. Select [L-HH] under ndv_l.syn and click OK. An Occurrence
Texture Parameters dialog appears.
3.Deselect all of the Textures to Compute options except for Data Range.
4.Set the Processing Window: Rows and Cols to 7 and 7.
5.In the Enter Output Filename field, enter ndv_hh.tex and click OK.
Create Color-coded Texture Map
6.In the Available Bands List, click Display #1 and select New Display.
7.Select Data Range under ndv_hh.tex and click Load Band.
8.From the Display #2 menu bar, select Enhance→[Image] Square Root.
9.From any Display group menu bar, select Tools→Link→ Link Displays. The Link Displays dialog appears.
Click OK to link the original image (Display #1) with the colored texture image (Display #2).
10.Click in an Image window to toggle between the two images.
11.Double-click inside an Image window to display the Cursor Location/Value tool. Examine the data values in the
textured image, and compare these to the original image.
12.From the Display #2 menu bar, select Tools→Color Mapping→Density Slice. A Density Slice Band Choice
dialog appears.
13.Select the Data Range band and click OK. A Density Slice dialog appears.
14.Accept the default ranges by clicking Apply.
15.Try creating your own density-sliced image and view the results.
16.Keep Display #2 open for the next exercise.
Image-Map Output
In this exercise, you will create a map of your color-coded textured image and add a border and map key.
1.From the Display #2 menu bar, select Overlay→Annotation. An Annotation dialog appears.
2.From the Annotation dialog menu bar, select Options→Set Display Borders.
3.In the Display Borders dialog, enter 100 in the upper field, and leave the remaining fields 0.
4.Click Border Color and select Items 1:20→White. Click OK. This adds a 100-pixel white border at the top of
the image.
5.Move the Image box in the Scroll window to the top of the image containing the border.
6.Enter a map title in the empty field in the Annotation dialog. Set the Size value to 16. Click the Color box once
to select black.
7.Click in the Image window to show the map title, then move it inside the white border to the far left. Right-click
to lock the map title in place. You can place multiple text items on the image in this manner, and you can change their font size, type, color, and thickness as desired.
8.From the Annotation dialog menu bar, select Object→Color Ramp.
9.Enter Min and Max values of 0 and 255 respectively, set Inc to 4, and set the font Size to 14 to annotate the
color ramp.
10.Click in the Image window to show the map key, move it inside the white border to the far right, then right-click
to lock it in place. The following figure shows a sample map; your results may be different.
11.Save the image to a PostScript file by selecting File→Save Image As→Postscript File from the Display #2
menu bar. An Output Display to PostScript File dialog appears.
12.Leave the default values, and enter an output filename or accept the default name of ndv_hh.ps. Click OK.
Or, output the map directly to your printer by selecting File → Print from the Display #2 menu bar.
13.When you are finished, select File→Exit from the ENVI main menu bar.
极化SAR影像分类综述
基于目标分解的极化SAR图像分类 摘要：极化SAR图像分类是SAR图像解译的重要内容，从现有的文献来看，基于目标分解理论的极化SAR图像分类算法是所有分类算法中较为实用、准确，且发展较快的。以此为研究背景，论文首先介绍了雷达极化的基础理论，并在此基础上系统地分析了当前各种典型目标分解算法的特性，最后总结了几种典型的基于目标分解理论的极化SAR图像分类算法。 关键词：极化SAR 目标分解图像分类 1引言 极化合成孔径雷达(SAR )通过测量地面每个分辨单元内的散射回波，进而获得其极化散射矩阵以及Stokes矩阵。极化散射矩阵将目标散射的能量特性、相位特性和极化特性统一起来，相对完整地描述了雷达目标的电磁散射特性，为更加深入地研究地物目标提供了重要的依据，极大地增强了成像雷达对目标信息的获取能力。 从极化SAR图像数据中，我们可以提取目标的极化散射特性，从而实现全极化数据的分类和聚类等其他应用。这需要我们对极化数据进行分析，有效地分离出目标的散射特性，其理论核心是目标分解。目标分解理论是Po1SA R图像处理技术中最基本的方法，目标分解的主要目的是把极化散射矩阵或相干矩阵和协方差矩阵分解成代表不同散射机理的若干项之和，每一项对应一定的物理意义。目标分解的突出优点就是它们大都具有明确的物理解释。因为目标回波的极化信息可以反映目标的几何结构和物理特性，所以极化目标分解理论可用于目标检测或分类。目前，极化目标分解理论主要分为基于散射矩阵分解的相干目标分解方法和基于协方差矩阵或相干矩阵的部分相干目标分解两类。本文从目标分解的基本理论出发，对这些分解方法进行了归纳和分析，以便对这些分解方法进行深刻的把握。为目标分解方法应用于SAR图像分类提供一些参考。 2 极化SAR图像的基本理论 2.1 极化合成孔径雷达概述 极化合成孔径雷达是合成孔径雷达向多功能方向发展的一个重要内容，它能
sar工具说明
sar ：（收集报告并保存） sar可以收集、报告、存储系统活动信息。具体分一下情况： ##当不指定interval参数，会全部显示以前收集到的文件内容。如果指定-f标志，sar从-f指定 的以前保存的文件中提取内容，然后写入标准输出，如果没有指定-f标志，将从默认的日报文 件（/var/log/sadd)读取。如果-f指定的文件或者默认位置没有文件，则提示没有那个文件目
录或目录错误。 ##当指定interval参数，如果指定了count则以interval为间隔显示count条，接下来说要显示 的数据来源：果指定-f标志，sar从-f指定的以前保存的文件中提取内容，然后写入标准输出 （-f -表示默认位置）；如果没有指定-f标志，将从系统收集信息并显示到控制台，如果指定了 -o标志，收集的数据同时会写到-o指定的文件中（-o -表示默认位置）和控制台。 如果没有-P标志，sar报告系统范围中整体cpu的使用的平均值；反之如果指定了-P标志，sar将只 报告某个cpu的使用状况；当-P ALL标志指定，每个单独的cpu和整体所有cpu的使用情况都被报告。 如果同时需要采样和统计报告，为sar指定输出文件将使得这很便利，运行如下命令：sar -o datafile interval count >/dev/null 2>&1 &
所有被捕获的数据会以二进制形式写入datafile中，这些数据可以用sar -f显示，如果不指定count参数，则文件 -b：io相关： tps ：每秒总共的物理设备的请求次数 rtps ：每秒总共的物理设备的写请求次数 wtps ：每秒的io写请求次数 bread/s ：每秒的io读请求扇区数（1扇区=512byte） bwrtn/s ：每秒的io写请求扇区数 -B:页相关信息 pgpgin/s ：每秒系统从磁盘置入的字节数（KB） pgpgout/s ：每秒系统置出到磁盘上的字节数（KB） fault/s ：每秒钟系统产生的页中断(major + minor)。 majflt/s ：每秒产生的主中断次数（见【注】）。 pgfree/s ：每秒被放入空闲队列中的页个数。 pgscank/s ：每秒被页面交换守护进程kswapd扫描的页个数。 pgscand/s ：每秒直接被扫描的页个数。
极化SAR影像分类综述
基于目标分解的极化SAR图像分类 硕研2010级6班金姗姗2010010615 摘要：极化SAR图像分类是SAR图像解译的重要内容，从现有的文献来看，基于目标分解理论的极化SAR图像分类算法是所有分类算法中较为实用、准确，且发展较快的。以此为研究背景，论文首先介绍了雷达极化的基础理论，并在此基础上系统地分析了当前各种典型目标分解算法的特性，最后总结了几种典型的基于目标分解理论的极化SAR图像分类算法。 关键词：极化SAR 目标分解图像分类 Abstract：Polarimetric SAR image classification is pivotal in SAR image interpretation. According to current literature, the classification algorithm for polarimetric SAR image based on target decomposition theorems is the most practical and exact one with fast developing speed among all algorithms. Under this background of research, the basic theory on radar polarimetric is discussed at first in this paper. Then the characteristic of typical target decomposition algorithms is analyzed in detail. Finally, typical polarimetric SAR image classification based on target decomposition theorems are summarized. Key words：POLSAR Target Decomposition Image Classification 1引言 极化合成孔径雷达(SAR )通过测量地面每个分辨单元内的散射回波，进而获得其极化散射矩阵以及Stokes矩阵。极化散射矩阵将目标散射的能量特性、相位特性和极化特性统一起来，相对完整地描述了雷达目标的电磁散射特性，为更加深入地研究地物目标提供了重要的依据，极大地增强了成像雷达对目标信息
极化干涉SAR的研究现状与启示_吴一戎
第29卷第5期电子与信息学报Vol.29No.5 2007年5月Journal of Electronics & Information Technology May2007 极化干涉SAR的研究现状与启示 吴一戎洪文王彦平 (中国科学院电子学研究所微波成像技术国家级重点实验室北京 100080) 摘要：阐述极化与干涉结合的基本考虑，介绍极化干涉SAR相干最优和相干目标分解的基本思想，总结分析极化干涉SAR技术、典型星载极化SAR系统研制和极化干涉SAR应用的研究现状，以得到开展极化干涉SAR技术研究的启示。 关键词：极化干涉SAR；极化SAR；干涉SAR；SAR 中图分类号：TN958 文献标识码：A 文章编号：1009-5896(2007)05-1258-05 The Current Status and Implications of Polarimetric SAR Interferometry Wu Yi-rong Hong Wen Wang Yan-ping (National Key Lab of Microwave Imaging Technology, Institute of Electronics, Chinese Academy of Sciences, Beijing 100080, China) Abstract: In this paper, the basic factors of the techniques combining the polarimetric synthetic aperture radar (SAR) with interferometric SAR are considered firstly, and then the basic concepts of coherence optimization and target decomposition of polarimetric SAR interferometry are illustrated. The current status of polarimetric SAR interferometry technique, the developments of typical spaceborne polarimetric SAR systems and the applications of polarimetric SAR interferometry are summarized. Key words: Polarimetric SAR Interferometry; Polarimetric SAR; Interferometric SAR; SAR 1 引言 经过长年的发展，合成孔径雷达(SAR-Synthetic Aperture Radar)技术与系统从单波段、单极化已逐步发展到多波段、全极化SAR、干涉SAR 遥感[1]，最近几年出现的极化干涉SAR (POLINSAR-Polarimetric SAR Interferometry) 把SAR遥感应用推向高潮，期望实现从高分辨率定性成像到精确高分辨率定量测量的转变。 POLINSAR通过极化和干涉信息的有效组合，可以同时提取观测对象的空间三维结构特征信息和散射信息，为微波定量遥感、高精度数字高程信息和观测对象细微形变信息的提取提供了可能性。POLINSAR系统研制、数据处理技术和应用研究已成为国外SAR技术研究的热点。 本文通过对POLINSAR技术、典型星载极化SAR系统研制和POLINSAR应用的研究现状进行总结分析，以期得到开展POLINSAR技术研究的启示。 2 POLINSAR简介 2.1极化与干涉结合的基本考虑 极化SAR(POLSAR-Polarimetric SAR)测量可获得每一像元的全散射矩阵，并合成包括线性极化、圆极化及椭圆极化在内的多种极化散射信息。因此与常规SAR相比，在雷达目标探测[1]、识别、纹理特征的提取、目标方向、物质对称性和组分方面研究具有很大的改善。POLSAR对植被散射 2006-06-20收到，2006-12-29改回体的形状和方向具有较强的敏感性[2]。通过测量每一像元的全极化散射矩阵，有可能将复杂的地物散射过程分解为几种单一的散射过程[3]，并利用地物在不同极化状态下的极化散射信息为更准确地探测目标特征提供可能。全极化数据对遥感定量测量具有很大的应用价值和潜力，是遥感定量测量的重要研究方向。 INSAR主要用于获取地物的空间垂直结构信息[4]。通过该技术可以获取的两个重要的参数分别为干涉相位和相干系数。对于相位，它已广泛应用于DEM生成、地震/火山/冰川/地表沉降和海洋物理参数获取的研究中。近年来，INSAR获取的另一个重要参数——相干系数已被逐步认识并开始应用于地表特征的基础性分析和地表植被高度及生物量的反演研究中，是一个极具潜力的研究领域。SAR干涉技术在实现过程中隐含着这样一个假设，即假定图像中每个像素的信号回波是从固定高度的参考平面上的一个散射中心散射回来的，因而测得的相位差就与这个参考平面的高度成正比。然而，由于地面坡度、粗糙度等因素的存在以及地表植被和体散射的影响，地物对电磁波散射的实际过程极为复杂，分辨单元内往往同时存在多种散射机理，且不同散射的相位中心亦可能位于不同高度上。这时，两幅干涉图之间的相位差只可能反映所有散射体的平均高度而无法反映某一特定散射中心的实际高度。例如，这种现象在有植被覆盖的区域显得尤为严重，这是因为植被覆盖区域的后向散射主要由植被层本身和由植被覆盖下的地面所散射的两种主要信号分量组成。这两个散射中心的高度通常是不同的，因此
基于ENVI的SAR数据处理流程介绍
ENVI Tutorial: Basic SAR Processing and Analysis Table of Contents O VERVIEW OF T HIS T UTORIAL (2) Background (2) S INGLE-B AND SAR P ROCESSING (3) Read and Display RADARSAT CEOS Data (3) Review CEOS Header (3) Apply Square-Root Contrast Stretch (4) Remove Speckle using Adaptive Filters (5) Density Slice (6) Edge Enhancement (7) Data Fusion (8) Image-Map Output (9)
Overview of This Tutorial This tutorial is designed to give you a working knowledge of ENVI’s basic tools for processing single-band synthetic aperture radar (SAR) data such as RADARSAT, ERS-1, and JERS-1. Files Used in This Tutorial ENVI Resource DVD: envidata\rsat_sub File Description lea_01.001 RADARSAT leader file bonnrsat.img (.hdr) RADARSAT image subset rsi_f1.img (.hdr) Frost filter result dslice.dsr Density slice file rsi_f2.img (.hdr) Laplacian filter result rsi_f3.img (.hdr) Laplacian filter result with 90% add-back rsi_fus.img (.hdr) Simulated fused TM and RADARSAT rsi_map.jpg RADARSAT map composition example Background Use the Radar menu in ENVI to access standard and advanced tools for analysis of detected radar images and advanced SAR systems such as NASA/Jet Propulsion Laboratory's (JPL's) fully polarimetric AIRSAR and SIR-C systems. ENVI can process ERS-1, JERS-1, RADARSAT, SIR-C, X-SAR, and AIRSAR data and any other detected SAR dataset. In addition, ENVI is designed to handle radar data distributed in the CEOS format. Most standard ENVI processing functions are inherently radar-capable, including all display capabilities, stretching, color manipulations, classification, registration, filters, geometric rectification, and so on. Additional specialized tools are provided for analyzing polarimetric radar data. A typical processing flow may include reviewing the CEOS header, reading the CEOS data, displaying and contrast stretching, removing speckle using an adaptive filter, density slicing, edge enhancement, data fusion, and map composition.
使用Cameron分解的极化SAR特征检测
使用Cameron分解的极化SAR特征检测 摘要—本文提供了一种检测特定的极化SAR特征的方法。一个分辨率单元的极化响应可以看做是这个分辨率单元的电磁散射矩阵的一个样本。尽管使用了多个相干孔径，还是能获得散射矩阵的多个样本。通过使用合适的分解和加权对数似然方程，估计被观察的散射矩阵响应匹配已知电磁特征的相似度是有可能的。 关键词：相干，多角度，极化，散射矩阵分解，子孔径处理。 I.引言 有多种分辨率，频率带宽和极化方式的SAR传感器有很大的应用范围，包括土地表面覆盖物的的表征和分类[1]-[5],[30]，和人造目标的探测/表征/分类，比如城市结构[6]-[8]，失事飞机[9]-[12]，轮船[13],[14]，军用车辆[15],[16]，地雷[17],[18]，未爆炸武器(UXO)[19]，等等。 全极化SAR数据集提供给了一个强有力的工具，可以用来描述允许使用复杂的散射矩阵响应的目标的散射行为。利用极化SAR成像的标准方法大多包括用分解变换(举例子来说，见[20]-[30])来使数据标准化这一过程。 极化分解一般分为明显的两类——相干和非相干分解。非相干分解(例如，Cloude-Plottier[3],[26])包含基于局部极化行为的统计信息，而且通常利用相关联的像素或分辨率单元的平均相关矩阵（或米勒矩阵）。相干分解（例如，Krogager[22],[23]，Cameron[24],[25]）是基于单个像素的响应，因此直接用于测量每个像素的复散射矩阵。因为在本文中我们关注的是特定散射中心的检测，不管周围像素的散射机制，所以我们不能使用任何一个利用局部统计信息的分解方法，这将限制我们对相干分解的讨论。 相干分解一般只能产生，由单个主散射中心组成的这些分辨率单元的有效的结果。在主散射体缺少时，斑点响应将导致一个明显的随机散射矩阵，而且产生一个随机响应。事实上，相干解调由于其固有的敏感的斑点噪声，在文献[26]中已经受到了批评。 相干方法的初始应用是描述以像素基础构成像素散射机制的基础的特征。有以下两点考虑是比较中肯的，首先，光极限散射的假设隐含了上散射类型的定义。第二，结果仅在一个给定的分辨率单元有占主要地位的单个散射中心时有效。类似的，正如也将在这篇文章中看到的，没有光极限假设会导致对单个主散射中心
SAR图像
合成孔径雷达 SAR是一种可成像的雷达，它所用的雷达波段大约是300MHz到30GHz。比如一般用的波段是1~10GHz的合成孔径雷达，大气对这种波段的影响不大。也就是说如果天上有一个合成孔径雷达卫星，白天黑夜、大气的云雾雨雪等天气变化对雷达看到的结果影响甚微，可忽略不计。所以合成孔径雷达是一种全天时、全天候的雷达，它所成的图像就是SAR图像了。 SAR图像的场景和照相机拍出来的场景类似，只不过波段不同看到的事物也不一样。SAR都是斜视的，而光学的可以垂直照射。 SAR卫星方面，我记得最早发射的是加拿大的Radarsat，且有后续计划。美国有航天飞机上载的SIR-C等合成孔径雷达。日本现在有ALOS卫星上载的PALSAR合成孔径雷达（1.27GHz）。德国的有TerraSAR系列。据我所知，现在分辨率最高的是德国的X波段SAR系统，数据不好弄到；日本的PALSAR 的SAR图像可以到官方网站下载到示例数据。加拿大的Radarsat和美国的SIR-C数据也是可以到网上下载到的。 机载SAR方面，几乎数的上的雷达强国都有自己的系统。机载SAR图像有日本的，法国的，德国的，美国的，但是在网络上找这种图像要费点功夫，不是很容易！中国虽然也有，但公开的资料较少，公开的图像资料就更少了。 SAR图像处理软件推荐欧空局的一个免费开源关键PolSARpro（我用过的），可以到欧空局网站上下载，里面的pdf有更详细的介绍SAR及其图像处理等内容。 inSAR技术 基于Photoshop插件架构的合成孔径雷达(SAR)图像处理与评估系统主要功能.以图像评估插件的开发为例对关键技术进行了分析.结果表明,采用Photoshop插件方式,可以避免复杂的内存管理编程和用户界面设计,充分利用Photoshop的图形处理功能,减少了工作量,并提高系统稳定性和可用性. 所以sar是基于photoshop插件的合成孔径雷达SAR(Synthetic Aperture Radar，合成孔径雷达) 问：机载合成孔径雷达sar 多久能生一幅图像就是说采图周期怎么算？ 最佳答案 我们按照最简单的条带式来说，每次生成的是一块图像，这块图像在距离向是全部的范围，在方位向则需要根据你系统的运算能力选择合适的长度。 在SAR中没有所谓采图周期这样的概念，采样是一直运行的，数据处理则在后台运行，把要成像的区域向前后各扩展半个孔径，然后做距离压缩方位压缩即可。当然这是没有考虑距离徙动校正的情况，如果考虑校正，则要选择算法，一般工程上就是RD的改进和CS。有什么问题可以联系liuchencalay@https://www.wendangku.net/doc/5318552743.html, 我国有机载合成孔径雷达吗？有的话性能如何？其他机载对地雷达如何？能被有
SAR数据价格
SAR数据价格 1、欧空局的ENVISAT ASAR数据（C波段） 数据处理系统在生成ASAR Level 1B产品时，可以运用不同的处理方式，从而得到不同的ASAR Level 1B产品供用户选择。 Image模式、Alternating Polarisation模式 这两种模式的Level 1B产品分为三类：Precision Image、Single Look Complex Image和Medium Resolution。 Precision Image是多视、地距图象，产品象元尺寸为12.5米，适合于大多数的应用。 Single Look Complex （SLC）Image是单视复型产品，产品的象元尺寸由成象的模式决定，可被用于SAR图象质量评估、标定和干涉、或风/海浪应用。在处理中较少对数据进行修正，以允许用户可以更自由地将数据处理为其它产品。 Medium Resolution是象元尺寸为75米的图象产品，产品的其它特性同Precision Image。Wide Swath模式 Wide Swath模式只提供象元尺寸为75米的Medium Resolution图象产品，在产品注解中也提供了完整的标定参数。 各种工作模式的特性见下表： ENVISAT数据产品的价格（不包括申请费）： Image Mode 和Wide Swath Mode :3000元/景 Alternating Polarisation Mode :4000元/景 我站ENVISAT数据编程申请的收费标准如下： a.提前14 天以上的编程申请为普通编程，不收申请费用； b.7-14 天编程申请为加急编程，每一数据段用户需付申请费3000元； c.2-7天编程申请为特急编程，每一数据段用户需付申请费8000元； （注：编程申请最少需要提前两天提交。） 2、RADARSAT-1数据（C波段）
基于ENVI的极化SAR数据处理流程介绍
ENVI Tutorial: Polarimetric SAR Processing and Analysis Table of Contents O VERVIEW OF T HIS T UTORIAL (2) Background: SIR-C/SAR (2) P REPARE SIR-C D ATA (3) Optional: Read a SIR-C CEOS Data Tape (3) Optional: Multilook SIR-C Data (3) S YNTHESIZE I MAGES (4) Default Polarization Combinations (4) Other Polarization Combinations (4) Display Images (5) Define ROIs for Polarization Signatures (6) Extract Polarization Signatures (6) Adaptive Filters (8) Slant-to-Ground Range Transformation (9) Preview CEOS Header (9) Resample Image (9) Texture Analysis (10) Create Color-coded Texture Map (10) Image-Map Output (11)
Overview of This Tutorial This tutorial demonstrates the use of ENVI’s tools for analyzing polarimetric synthetic aperture radar (SAR) data. You will multilook Spaceborne Imaging Radar-C (SIR-C) from Death Valley, California; synthesize images, define ROIs for (and extract) polarization signatures, use adaptive filters, perform slant-to-ground range transformation, use texture analysis, and create an output image-map. Files Used in This Tutorial ENVI Tutorial Data DVD: envidata\ndv_sirc File Description ndv_l.cdp L-band SIR-C subset in ENVI compressed data product (.cdp) format pol_sig.roi Region of interest (ROI) file Background: SIR-C/SAR SIR-C is a polarimetric SAR instrument that uses two microwave wavelengths: L-band (24 cm) and C-band (6 cm). The SIR-C radar system was flown as a science experiment on the Space Shuttle Endeavor in April (SRL-1) and October 1994 (SRL-2), collecting high-quality SAR data over many sites around the world. (A second radar system, XSAR, was also flown on this mission, but these data are neither discussed nor processed here.) Additional information about SIR-C is available on the NASA/JPL Imaging Radar Home Page at https://www.wendangku.net/doc/5318552743.html,/.
常用卫星数据介绍
国外卫星有： WorldView 1/2/3，GeoEye1/2，RapidEye,IKONOS,QuickBird,Spot5,Spot6,Landsat-5 TM,Landsat-7 ETM+,Landsat-8 ALI,Pleiades,Alos,terrasar-x,radarsat-2,全美锁眼卫星全系列(1960-1980)，印度Cartosat-1(又名IRA-P5) 国内卫星有: HJ-A/B CCD,ZY-02-C,ZY-3,CBERS-3/4,天绘系统,高分系列,资源系列等 一、Landsat7卫星的TM/ETM+数据介绍 TM是一种遥感器，搭载在美国陆地卫星Landsat系列卫星上。TM影像是指美国陆地卫星4～5号专题制图仪（thematic mapper）所获取的多波段扫描影像。有7个波段 Landsat-7，星上携带专题制图仪ETM，ETM具有8个波段，其中1-5波段和7波段是多光谱波段，空间分辨率是30米，第六波段是热红外波段，空间分辨率是120米，第8波段为全色波段，分辨率为15米。景宽185公里，景面积为34225平方公里。 波段介绍： 1.TM1 0.45-0.52um,蓝波段 对水体穿透强, 该波段位于水体衰减系数最小，散射最弱的部位（0.45— 0.55um）,对水体的穿透力最大，可获得更多水下信息，用于判断水深，浅海水下地形，水体浑浊度，沿岸水，地表水等；能够反射浅水水下特征，区分土壤和植被、编制森林类型图、区分人造地物类型,分析土地利用。对叶绿素与叶色素反映敏感,有助于判别水深及水中叶绿素分布以及水中是否有水华等。 2.TM2 0.52-0.60um,绿波段 对植物的绿反射敏感该波段位于健康绿色植物的绿色反射率（0.54—-0.55um）附近；对健康茂盛植物的反射敏感, 主要观测植被在绿波段中的反射峰值,这一波段位于叶绿素的两个吸收带之间,利用这一波段增强鉴别植被的能力对绿的穿透力强, 探测健康植被绿色反射率,按绿峰反射评价植物的生活状况,区分林型,树种，植被类型和评估作物长势对水体有一定的穿透力，可反映水下特征，水体浑浊度，水下地形，沙洲，沿岸沙地等。. 可区分人造地物类型， 3.TM3 0.62-0.69um ,红波段 对水中悬浮泥沙反映敏感。该波段位于含沙浓度不同的水体辐射峰值（0.58—-0.68um）附近，对水中悬浮泥沙反映敏感。叶绿素的主要吸收波段, 能增强植被覆盖与无植被覆盖之间的反差,亦能增强同类植被的反差，反映不同植物叶绿素吸收,植物健康状况,用于区分植物种类与植物覆盖率, 测量植物绿色素吸收率，并以此进行植物分类；此外其信息量大,广泛用于对裸露地表，植被，岩性，地层，构造，地貌等为可见光最佳波段；可区分人造地物类型
ENVI对SAR数据的预处理过程(详细版)资料
E N V I对S A R数据的预处理过程(详细版)
一、数据的导入： (1) 在 Toolbox 中，选择 SARscape ->Basic->Import Data->Standard Formats- >ALOS PALSAR。 (2) 在打开的面板中，数据类型（Data Type）：JAXA-FBD Level 1.1。 注：这些信息可以从数据文件名中推导而来。 (3) 单击 Leader/Param file，选择 d1300816-005-ALPSRP246750820-H1.1__A\LED-ALPSRP246750820-H1.1__A文件。 (4) 点击 Data list，选择 d1300816-005-ALPSRP246750820-H1.1__A\IMG-HH-ALPSRP246750820- H1.1__A文件 (4) 单击 Output file，选择输出路径。 注：软件会在输入文件名的基础上增加几个标识字母，如这里增加“_SLC”(5) 单击 Start 执行，最后输出结果是 ENVI 的slc文件，sml格式的元数据文件，hdr格式的头文件等。 (6) 可在 ENVI 中打开导入生成的以slc为后缀的 SAR 图像文件。
二、多视 单视复数（SLC）SAR 图像产品包含很多的斑点噪声，为了得到最高空间分辨率的 SAR图像，SAR 信号处理器使用完整的合成孔径和所有的信号数据。多视处理是在图像的距离向和方位向上的分辨率做了平均，目的是为了抑制 SAR 图像的斑点噪声。多视的图像提高了辐射分辨率，降低了空间分辨率。 (1) 在 Toolbox 中，选择 SARscape->Basic ->Multilooking。 (2) 单击 Input file 按钮，选择一景 SLC 数据（前面导入生成的 ALOS PALSAR 数据）。 注意：文件选择框的文件类型默认是*_slc，就是文件名以_slc 结尾的文件，如不是，可选择*.*。 (3) 设置：方位向视数（Azimuth Looks）：5，距离向视数（Range Looks）：1 注：详细的计算方法如下所述。另外，单击 Look 按钮可以估算视数。
SAR数据介绍
Zondy SAR数据介绍 本文对当前主要的SAR卫星和对应的数据做了一定的介绍，并且对当前平台上有的数据进行了一定的整理，不足之处希望修改。 Writer：Huang Xiaodong Date：Jul-26-2010 Email：rs.hxd@https://www.wendangku.net/doc/5318552743.html,
目录 ALOS (4) 卫星介绍 (4) 数据格式 (5) 主要用途 (6) 官方网址 (6) 现有数据 (6) ERS1/2 (6) 卫星介绍 (6) 数据格式（CEOS） (7) 主要用途 (7) 官方网站 (7) 现有数据 (7) Radarsat 1 (8) 卫星介绍 (8) 工作模式 (8) 数据格式（CEOS） (9) 主要用途 (9) 官方网站 (9) 现有数据 (9) Radarsat 2 (9) 卫星介绍 (9) 工作模式 (10) 数据格式（*.tif） (10) 主要用途 (10) 官方网站 (10) 现有数据 (10) Envisat-1 (11) 卫星介绍 (11) ASAR工作模式 (11) ASAR产品介绍 (12) Level 0 产品 (12) Level 1B产品 (13) 数据格式（*.N1） (13) 主要用途 (14) 官方网站 (15) 现有数据 (15) TerraSAR-X and TanDEM-X (16) 卫星介绍 (16) 工作模式 (17) 数据格式（SLC:*.cos;Other:*.tif） (17) 主要用途 (18)
官方网站 (18) 现有数据 (18) COSMO-SkyMed (19) 卫星介绍 (19) 成像模式 (20) 数据格式（*.HDF5） (20) 主要用途 (20) 官方网站 (20) 现有数据 (21) JERS (21) 卫星介绍 (21) 数据格式（CEOS） (22) 主要用途 (22) 官方网站 (22) 现有数据 (22)
不同分解方法的极化SAR数据分类
由于平时分类的时候，一般采用的是Cloude 分解的得到的特征值进行分类的。而对于不同的数据，分解方式的不同，相应的分类结果也存在一定的差异，而且对于不同类型的数据，其最优分解方法并不一定是Cloude 分解。所以对两组数据（旧金山数据和海南数据）进行基于不同分解方法的分类实验，对比不同分解下的效果。 这部分实验首先对数据进行极化分解，然后利用分解得到的特征量进行分类。共进行了基于Cloude ，Freeman2，Freeman3，Krogager ，Vanzel ，Yamaguchi3，Yamaguchi4，7种特征分解的分类实验，分类算法采用的基于Wishart 分布的最大似然分类器。 旧金山数据的7种分解方法的分类结果 Cloude 是用的最多的分解方法，除了把数据分解为与散射机制有关的三个特征值321,,λλλ（分别代表三种散射机制：平面散射，二面角散射和体散射），还有具有旋转不变性的散射角a 和熵H 。所以基于cloude 分解的分类结果较为平稳。而对于其他6种分解方式，都是把数据分解为与某种散射机制类型对应的参数，代表该散射类型的强度，不具有旋转不变性的参数，所以适用数据的类型并不是很广，所以不像Cloude 分解那样对各种类型的数据都适用。从下面7幅分类结果看到，基于Cloude 分解的分类结果整体分类效果较好，不存在大范围的错误分类区。但是其他6种分解方式的分类结果，海洋均被划分为多个层次，所以这6种分解方式对海洋的分类适用性不好。但是对比7个分类结果发现，每种分解的分类结果各有优势，如Yamaguchi4和Freeman2中植被和城区的划分效果较好，马球场和高尔夫球场结构完整，而且城区内部道路细节分明，效果均优于Cloude 分解的分类结果。 图1(a)基于Cloude 分解的分类结果 图1(b)基于Vanzel 分解的分类结果
船舶SAR图像数据集简介
船舶SAR图像数据集简介 深度学习技术的发展和计算能力的提高极大地促进了SAR图像数据集的建立和数据集规模的提高，本节将介绍舰船领域三个重要的SAR图像数据集，SSDD数据集，HRSID数据集和Yuanyuan Wang等人建立的复杂背景船舶数据集。这三个数据集为深度学习和计算机视觉技术应用于SAR图像目标检测起到了基础性的作用。其中，本文提出的工作用到了SSDD数据集和HRSID数据集。 首先介绍SSDD数据集[11]，SSDD数据集是国内外第一个专门用于基于SAR 图像的舰船目标检测公开数据集，数据集包含各种情况下的船舶图像，如不同图像分辨率、船舶尺寸、海况、传感器类型等，可以作为研究人员评估其算法的基准。对于SSDD数据集中的每一艘船，都标注有带置信分数的边界框。由于该数据集的应用范围大多是视觉对象检测，因此其构建方法类似于PASCAL VOC数据集[12]。SSDD数据集由三个子集组成，包括训练集、验证集和测试集，各个部分图像数量的比例为7:2:1。由于SSDD数据集包含的不同条件如表1中所示，例如不同的图像分辨率，图像大小，海况，传感器等等。因此，这种数据集设置可以使得训练出来的目标检测器更加具有鲁棒性，但是这也会使得目标检测器很难在该数据集上获得非常高的性能。 SSDD数据集中船只和图像的数量统计如表2所示，其中NoS表示船舶数量，NoI表示图像数量。在SSDD数据集中，总共有1160张图片和2456艘船。每幅图像的平均船舶数量为2.12艘。在使用该数据集的过程中可以根据所选算法的要求对数据集进行扩展。尽管SSDD数据集的规模不及PASCAL VOC数据集，但是SSDD数据集足够用来测试基于目标检测任务的算法性能，因此可以通过结合防止过拟合的技巧，比如正则化，来训练一个目标检测模型。本文利用开源的“labelimg” 软件制作标签，每个船的边框会被表示成（x, y, w, h）。这里（x, y）是矩形中心点的坐标，w 是矩形的宽度，h 是矩形的高度。 船舶，图(b)显示远海的船舶，图(c)显示复杂背景的船舶
ENVI对SAR数据的预处理过程(详细版)
一、数据的导入： (1) 在Toolbox 中，选择SARscape ->Basic->Import Data->Standard Formats->ALOS PALSAR。 (2) 在打开的面板中，数据类型（Data Type）：JAXA-FBD Level 1.1。 注：这些信息可以从数据文件名中推导而来。 (3) 单击Leader/Param file，选择 d1300816-005-ALPSRP246750820-H1.1__A\LED-ALPSRP246750820-H1.1__A文件。 (4) 点击Data list，选择 d1300816-005-ALPSRP246750820-H1.1__A\IMG-HH-ALPSRP246750820-H1.1__A文件 (4) 单击Output file，选择输出路径。 注：软件会在输入文件名的基础上增加几个标识字母，如这里增加“_SLC” (5) 单击Start 执行，最后输出结果是ENVI 的slc文件，sml格式的元数据文件，hdr格式的头文件等。 (6) 可在ENVI 中打开导入生成的以slc为后缀的SAR 图像文件。 二、多视 单视复数（SLC）SAR 图像产品包含很多的斑点噪声，为了得到最高空间分辨率的SAR图像，SAR 信号处理器使用完整的合成孔径和所有的信号数据。多视处理是在图像的距离向和方位向上的分辨率做了平均，目的是为了抑制SAR 图像的斑点噪声。多视的图像提高了辐射分辨率，降低了空间分辨率。 (1) 在Toolbox 中，选择SARscape->Basic ->Multilooking。 (2) 单击Input file 按钮，选择一景SLC 数据（前面导入生成的ALOS PALSAR数据）。 注意：文件选择框的文件类型默认是*_slc，就是文件名以_slc 结尾的文件，如不是，可选
SAR卫星数据
表1-1 已发射的SAR卫星 Tab.1-1 SAR satellites have launched[4] 发射者星载SAR 发射时间 美国Seasat SIR-A SIR-B SIR-C Light SAR 1978.06 1981.11(航天飞机) 1984.10(航天飞机) 1994.09(航天飞机) 2002.9 俄罗斯KOSMOS 1870 A1 maz-1 A1 maz-1A A1 maz-1B A1 maz-2 1987 1991.3 1993 1997 2004 ESA(欧空局) ERS-1 ERS-2 Envisat-1 1991.7 1995.4 2002.3 日本JERS-1 ALOS 1992.2 2006.1 加拿大Radarsat-1 Radarsat-1 1995.11 2007.12 德国TerraSAR-X 2007.6 意大利COSMO-Sky Med 2007.6
表1-2 已发射的SAR卫星的主要技术性能 Tab.1-2 The main technical performance of the SAR satellites have launched[4] 卫星型号Seasat-1 SIR-A SIR-B SIR-C/X-SAR Lacrosse5 Radarsat-1 Radarsat- 2 国家美国美国美国美国/意大利 和德国合作 美国加拿大加拿大 运行时间1978.6 1981.11 1984.10 1994.4.9 1994.9 2000.2.11 2005.4 1995.11 2007.12 轨道高度/km 805 259 225 225 235（IFSAR） 718 798 798 工作频率/GHz L（1.2795）L（1.275）L（1.275）L（1.275） C（5.1） X（9.65） C X C（5.3）C（5.3） 工作波长/cm 23.5 23.5 23.5 23.5，5.8， 3.1 5.66 5.66 极化HH HH HH HH、HV、 VH、VV HH HH、HV、 VH、VV 空间分辨率/m 25×25 40×40 （20～50）×25 25×12.5（L、C） 10/25（X） 6-16（IFSAR） 0.3(精扫) 1（标准） 3（宽扫） 25×28 9×9 50×50/ 100×100 （30～45） ×28 3～100