Coastline Detection Method in SAR Image Based on Improved Contourlet Transform

Abstract

The present invention discloses a SAR image coastline detection method based on improved Contourlet transform, which solves a basic problem of marine resource management - SAR image coastline detection. The present invention adopts a hyperanalysis wavelet transform with the same redundancy and simple structure to replace the Laplace pyramid transform in the Contourlet transform process to solve the translation invariance, and adopts a two-dimensional Gaussian low-pass filter with positive coefficients to reduce the fluctuation formed after filtering, thereby weakening the pseudo contour. On the other hand, the downsampling operation will bring frequency mixing to the image, thereby causing image distortion. The present invention combines double-pass Gaussian filtering to form a mother wavelet as a complex wavelet HWT, adopts a non-downsampled directional filter to improve the Contourlet transform for edge detection, and combines the DS evidence theory to judge multi-scale edges, and uses it for SAR image coastline detection.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, in particular to a SAR image coastline detection method based on improved Contourlet transform. Background Art

[0002] Rapid and accurate determination of coastlines is of great significance for the management of sea area use. Traditional field survey methods are labor-intensive, inefficient, and have long working cycles, and the data obtained are not easy to count. Remote sensing has a strong data acquisition capability and has obvious advantages in coastline surveys. Visible light remote sensing is often used for coastline detection because its imaging characteristics are consistent with human visual characteristics, but it is greatly restricted by lighting and climatic conditions. Compared with visible light remote sensing, synthetic aperture radar (SAR) has the characteristics of working all day and all weather, can image a large area, and can record information on coastline changes under severe weather conditions. It has been well applied in coastline detection. However, due to the influence of factors such as wind, waves, and speckle effects, the contrast between land and sea is not strong and the boundaries are not clear, which makes coastline detection in SAR images a major problem. Existing SAR image coastline detection technologies mostly use iterative calculations based on active contour models or level sets, which have high computational complexity and the detection accuracy is greatly affected by factors such as initial contours and window size.

[0003] The multi-scale geometric analysis method has multi-directional and anisotropic characteristics, and can effectively capture the high-dimensional anisotropy in the image. It can sparsely represent the image. After multi-scale decomposition, the energy of the image is more concentrated, which is more conducive to tracking the geometric features in the image. Sparse representation of SAR images will help improve the speed of coastline detection. The Contourlet transform in the multi-scale analysis method is directly obtained in the discrete domain. It is a "real" two-dimensional image representation method with very good nonlinear approximation performance. It can concentrate more coefficients into fewer Contourlet coefficients. Sparse representation of the image is conducive to quickly extracting SAR image contour information for further coastline detection. The transform has better nonlinear approximation ability and can approximate the optimal approximation of any two-dimensional piecewise function containing linear singularities. Based on the Contourlet transform and improving it, it will be beneficial to the rapid coastline detection of SAR images. On the other hand, the information fusion of the multi-scale edges obtained after the Contourlet transform using the DS evidence theory will help improve the accuracy of coastline detection. Summary of the invention

[0004] The purpose of the present invention is to improve the detection speed of coastline in SAR images and improve the detection accuracy of coastline. The key lies in reasonably representing SAR images to reduce the amount of calculation and reduce the influence of speckle noise to improve the credibility of the results.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A coastline detection method for SAR images based on improved Contourlet transform includes the following steps:

[0007] Step 1: Perform logarithmic transformation on the input SAR image I, convert the coherent noise into Gaussian noise, and obtain image I';

[0008] Step 2: Perform improved Contourlet decomposition on image I'. The specific process is as follows:

[0009] Process 1: Contourlet transform is usually implemented by two parts: Laplace pyramid transform and directional filter group. The coefficients of low-pass decomposition and reconstruction filters in the pyramid transform process are usually negative. After the isolated points pass through the filter, a small range of fluctuations will be formed, resulting in pseudo contours. If the coefficients of the low-pass decomposition and reconstruction filters can be made positive, the fluctuations formed after filtering will also be relatively small, thereby reducing pseudo contours. The two-dimensional Gaussian low-pass filter coefficients proposed by Burt can all be positive numbers, making it possible to eliminate pseudo contours by adjusting the positive and negative values ​​of the filter coefficients.

[0010] Preferably, a 5×5 two-dimensional Gaussian low-pass filter is used to perform low-pass decomposition on the SAR image, and the coefficient of the filter is: Where m, n = -2, -1, ..., 2, and h(n) is a one-dimensional filter with a length of 5: h(0) = α, h(-1) = h(1) = 1 / 4, h(-2) = h(2) = 1 / 4-α / 2. To ensure that the filter coefficients are all positive numbers, the parameter α is taken as 0.3.

[0011] Process 2: Contourlet performs downsampling operations in both the Laplace transform and the directional filter bank stages, which makes the transform lack translation invariance. To solve this problem, preferably, HWT is used to construct a two-dimensional hybrid transform with shift invariance:

[0012]

[0013]

[0014] Among them, (k1, k2) represents the position of the transformation, m represents the scale of decomposition, and are the Hyper-analysis wavelet transform (HWT) in the n1 and n2 directions respectively, and are the corresponding inverse transforms, H and H -1 is the one-dimensional Hilbert transform and its inverse transform. The HWT transform process uses the generating function obtained in process 1 to construct.

[0015] Process 3: Use a non-downsampled directional filter to perform directional filtering on the obtained HWT transform coefficients to obtain the directional subband space. The filter bank construction uses the complementary sector filter as the basic module, on which upsampling and linear transformation are performed to obtain support characteristics in different directions. The sector filter is downsampled by matrix Q and R i (i=0,1,2,3) upsampling operation, Q and R i They are:

[0016]

[0017] Step 3: Perform Principal Component Analysis (PCA) threshold processing on the high-frequency sub-band, and retain the principal component of the low-frequency sub-band using PCA;

[0018] Step 4: Calculate the candidate edge points of each layer based on the Canny criterion, and perform double threshold selection on the candidate points;

[0019] Step 5: The multi-scale edge of the SAR image is obtained by applying the DS evidence theory to the candidate edge points of each scale, and the final SAR image coastline detection result is obtained.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] 1. The coastline detection speed is improved by sparsely representing SAR images. The hyperanalytic wavelet transform HWT with the same redundancy and simple structure is used to replace the Laplace pyramid LP transform in the Contourlet transform process to solve the translation invariance, and a two-dimensional Gaussian low-pass filter with positive coefficients is used to reduce the fluctuation formed after filtering, thereby weakening the pseudo contour.

[0022] 2. By using a non-subsampled directional filter to improve the Contourlet transform for edge detection and combining it with the DS evidence theory to determine multi-scale edges, it is beneficial to improve the accuracy of coastline detection in SAR images. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a specific flow chart of the present invention. DETAILED DESCRIPTION

[0024] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The embodiments are only used to explain the present invention and do not limit the protection scope of the present invention.

[0025] Embodiment: A SAR image coastline detection method based on improved Contourlet transform, the implementation process is as follows Figure 1 As shown, the specific steps are:

[0026] Step 1: Perform logarithmic transformation on the input SAR image I, convert the coherent noise into Gaussian noise, and obtain image I';

[0027] Step 2: Perform improved Contourlet decomposition on image I';

[0028] A 5×5 two-dimensional Gaussian low-pass filter is used to perform low-pass decomposition on the SAR image. The coefficient of the filter is: Where m, n = -2, -1, ..., 2, and h(n) is a one-dimensional filter with a length of 5: h(0) = α, h(-1) = h(1) = 1 / 4, h(-2) = h(2) = 1 / 4-α / 2. To ensure that the filter coefficients are all positive numbers, the parameter α is taken as 0.3.

[0029] Use HWT to construct a two-dimensional hybrid transform with shift invariance:

[0030]

[0031]

[0032] Among them, (k1, k2) represents the position of the transformation, m represents the scale of decomposition, and are the HWT in the n1 and n2 directions respectively, and are the corresponding inverse transforms, H and H -1 is the one-dimensional Hilbert transform and its inverse transform. The HWT transform process is constructed using the generating function obtained in step 1.

[0033] The obtained HWT transform coefficients are directional filtered using a non-downsampled directional filter to obtain the directional subband space. The filter bank is constructed with a complementary sector filter as the basic module, on which upsampling and linear transformation are performed to obtain support characteristics in different directions. The sector filter is downsampled by matrix Q and R i (i=0,1,2,3) upsampling operation, Q and R i They are:

[0034]

[0035] Step 3: Perform PCA threshold processing on the high-frequency sub-band, and use PCA to retain the principal components of the low-frequency sub-band;

[0036] Step 4: Calculate the candidate edge points of each layer based on the Canny criterion, and perform double threshold selection on the candidate points;

[0037] Step 5: The multi-scale edge of the SAR image is obtained by applying the DS evidence theory to the candidate edge points of each scale, and the final SAR image coastline detection result is obtained.

[0038] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A coastline detection method for SAR images based on improved Contourlet transform, characterized in that: The following steps are involved: Step 1: Perform logarithmic transformation on the input SAR image I, convert the coherent noise into Gaussian noise, and obtain image I'; Step 2: Perform improved Contourlet decomposition on image I'. The specific process is as follows: Process 1: Use a two-dimensional Gaussian low-pass filter to perform low-pass decomposition on the SAR image. Process 2: Use the hyperanalytical wavelet transform to construct a two-dimensional hybrid transform with shift invariance. Process 3: Use a non-subsampled directional filter to perform directional filtering on the obtained super-analysis wavelet transform coefficients to obtain the directional subband space. In the process 1, a 5×5 two-dimensional Gaussian low-pass filter is used to perform low-pass decomposition on the SAR image, and the coefficient of the filter is: where m, n = -2, -1, ..., 2, and h(m) and h(n) are one-dimensional filters of length 5: h(0) = α, To ensure that the filter coefficients are all positive, the parameter α is taken as 0.

3. In the process 2, the super-analytical wavelet transform is used to construct a two-dimensional hybrid transform with shift invariance, and the transformation equation is: Among them, (k1, k2) represents the position of the transformation, m represents the scale of decomposition, and are the hyper-analytical wavelet transforms in the n1 and n2 directions, and are the corresponding inverse transforms, H and H -1 is the one-dimensional Hilbert transform and its inverse transform, The filter bank structure in the process 3 is based on the complementary sector filter as the basic module, on which upsampling and linear transformation are performed to obtain support characteristics in different directions, and the sector filter is subjected to matrix Q downsampling and Ri, i = 0, 1, 2, 3 upsampling operations, where Q and Ri are: Step 3: Threshold processing is performed on the high-frequency sub-band, and PCA is used to retain the principal components of the low-frequency sub-band; Step 4: Calculate the candidate edge points of each layer based on the Canny criterion, and perform double threshold selection on the candidate edge points; Step 5: The multi-scale edge of the SAR image is obtained by applying the DS evidence theory to the candidate edge points of each scale, and the final SAR image coastline detection result is obtained.

2. The SAR image coastline detection method based on improved Contourlet transform according to claim 1, characterized in that: In the step 5, the multi-scale candidate edge points are fused using the DS evidence theory to obtain the final SAR image coastline detection result.

Images