Spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT

Abstract

The invention provides a spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT. The method is characterized by comprising the following steps: S1, acquiring an interferogram through a spatial heterodyne spectrometer, calculating edge response by using a Prewitt operator, and extracting edge features; and S2, on the basis of a Gaussian filter, constructing a bilateral filter combining a gray domain weight and a spatial domain weight, and improving the denoising effect of bilateral filtering. And S3, performing non-subsampled wavelet transform (UDWT) on the processed interferogram, decomposing the processed interferogram into sub-bands of multiple scales and directions, and performing reconstruction to obtain a denoised interferogram. S4, spectrum recovery is achieved through Fourier transform, a denoised spectrum signal is obtained, the characteristic peak of the denoised spectrum signal is more obvious, the curve is smoother, and meanwhile the peak signal-to-noise ratio and non-uniformity are remarkably improved. The method has a remarkable noise removal effect on spectral signals, and can be widely applied to noise processing of spatial heterodyne spectrums.

Description

(I) Technical field

[0001] The invention relates to a spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT, which can correct spatial heterodyne interference data containing noise and belongs to the technical field of image processing. (II) Background technology

[0002] Spatial heterodyne interferometry is a high-resolution spectral detection technology based on the improvement of Michelson interferometer, which is mainly used in atmospheric monitoring, satellite remote sensing, chemical analysis and other fields. This technology combines traditional interference technology and spatial modulation technology, and has the advantages of ultra-high spectral resolution, high light flux, real-time and compact structure. The core working principle of spatial heterodyne interferometry is to obtain spectral information through spatial modulation of interference fringes, and no moving parts are required, which improves the stability and accuracy of the system. In actual measurements, spatial heterodyne spectrometers usually contain some noise due to instrument or environmental reasons, and appropriate algorithms need to be used for correction processing.

[0003] Interference pattern denoising is a common technical challenge in optical measurement, remote sensing and other fields. Since the interference pattern contains a lot of phase information and fringe structure, the presence of noise will seriously affect the quality and accuracy of the image. To this end, researchers have proposed a variety of denoising methods, such as using traditional methods such as mean filtering and Gaussian filtering to remove noise by smoothing the image in the spatial domain. Although these methods are computationally simple, they blur the edges and details of the interference pattern, especially when dealing with interference patterns with subtle fringes and phase changes, the effect is limited. Bilateral filtering combines the weighted average of spatial distance and pixel grayscale difference, which can retain edge information while removing noise. Although it avoids the edge blurring problem of traditional filters to a certain extent, due to the fixed parameter limitation, bilateral filtering still has shortcomings when dealing with interference patterns with complex edges or more details. Wavelet transform is a commonly used multi-scale analysis tool in image denoising. Through wavelet transform, the interference pattern can be decomposed into high-frequency and low-frequency sub-bands of multiple scales, and denoising is achieved by processing high-frequency noise components. However, the downsampling operation of traditional wavelet transform will lead to a reduction in image resolution and is not invariant to translation.

[0004] Based on the above background technology, the present invention proposes a spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT. By constructing a framework combining edge response guided adaptive bilateral filtering with non-subsampled wavelet transform, accurate noise suppression and detail preservation for complex interference patterns are achieved. Compared with traditional methods, this algorithm can not only efficiently denoise when processing images, but also retain the edge and detail features of the image, and is suitable for image processing tasks with high precision requirements. (III) Summary of the invention

[0005] The object of the present invention is to provide a spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT.

[0006] To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0007] Step S1, using the horizontal and vertical convolution kernels of the edge detection operator (Prewitt operator) to calculate the horizontal and vertical gradients of the image respectively, so as to calculate the edge response intensity of each pixel.

[0008] Step S2, adaptively adjust the bilateral filter parameters according to the edge response. Dynamically adjust the two main parameters of the bilateral filter according to the edge strength: spatial sum weight and grayscale kernel weight. Then, use the adaptively adjusted bilateral filter parameters to process the image.

[0009] Step S3, multi-scale decomposition of the image is performed, and sub-bands of different scales are decomposed by non-sub-sampling filter bank (NSP). At each scale, the high-frequency self-band is decomposed in multiple directions, and non-sub-sampling directional filter (NSDFB) is applied for directional filtering to capture the details and edge information of the image in different directions. By inverse filtering the sub-bands of each level, the detail sub-bands of each scale and direction are merged to reconstruct the high-frequency and low-frequency components. Reconstruction is performed step by step upward, and finally the denoised image is restored.

[0010] Step S4, performing a line-by-line fast Fourier transform on the denoised image to obtain spectral data after denoising. (IV) Description of the drawings

[0011] Figure 1 Flow chart of the spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT.

[0012] Figure 2 This is the ideal methane interference pattern.

[0013] Figure 3 Methane interference pattern with random noise.

[0014] Figure 4 This is an ideal methane spectrum point diagram.

[0015] Figure 5 This is a dot plot of the methane spectrum with random noise added.

[0016] Figure 6 This is the methane spectrum point map after only adaptive bilateral filtering denoising.

[0017] Figure 7 This is the methane spectrum point map after UDWT denoising only

[0018] Figure 8This is the methane spectrum point diagram after denoising by the spatial heterodyne interferometry correction method based on adaptive bilateral filtering and UDWT. (V) Specific implementation methods

[0019] The present invention will be further described below in conjunction with specific embodiments.

[0020] Step S1: simulate an ideal spatial heterodyne interference pattern based on the ideal methane spectrum, such as Figure 2 As shown. And assign random noise to it, the noise that can be assigned includes Gaussian noise and Poisson noise.

[0021] Step S2, using the Prewitt operator to dynamically adjust the bilateral filtering parameters, and perform adaptive bilateral filtering on the interference pattern. The specific steps are:

[0022] Use the edge detection operator (Prewitt operator) to calculate the edge response strength of each pixel. Among them, the Prewitt operator is a classic edge detection operator used to detect edges in images. Its basic principle is to determine the position and direction of the edge by calculating the local gradient of the grayscale value of the pixel in the image. It uses two convolution kernels to calculate the gradient in the horizontal and vertical directions respectively, and then combines the gradients in these two directions to determine the edge strength and direction of each pixel. Among them, The horizontal convolution kernel is: The vertical convolution kernel is: Using G x and G y Convolution operations are performed on the image to obtain the gradient maps of the image in the horizontal and vertical directions, which are respectively denoted as I and x and I y The gradient magnitude E(p) (edge ​​strength) of the image is:

[0023] Construct an adaptive bilateral filter. According to the edge strength E(p), dynamically adjust the parameters of the bilateral filter. In the edge area, in order to retain details, reduce the grayscale domain standard deviation, so that the filter is more sensitive to grayscale differences, thereby avoiding blur in the edge area. In the flat area, appropriately increase the grayscale domain standard deviation to enhance the smoothing effect and remove noise. The grayscale domain standard deviation parameter can be expressed as: in, is the initial grayscale kernel weight; E(p) is the edge strength at position p; E max is the maximum value of edge strength, used for normalization.

[0024] The interference image is filtered for the first time using the adaptively adjusted bilateral filtering parameters. The calculation formula for bilateral filtering is as follows: in,

[0025] After the adaptive bilateral filtering, although some noise in the interference pattern was initially removed, there were still some residual noises that were not completely removed due to the limitations of the bilateral filter in processing complex structures and details. Adaptive bilateral filtering mainly focuses on local smoothing, but in some areas with image edges and details, it may lead to insufficient detail retention or inability to completely eliminate some noise. Therefore, although the interference pattern after filtering has improved, it has not yet achieved the ideal denoising effect.

[0026] Step S3: To further improve the denoising effect, a two-dimensional non-subsampled wavelet transform filtering is performed on the interference image after the initial filtering. The specific steps are as follows.

[0027] The interference graph is decomposed into multi-scale sub-bands of different scales through the non-subsampling filter bank (NSP). The formula is expressed as: Among them, S j represents the low-frequency subband of level j, D j represents the high frequency subband of the jth level, and denote low-pass and high-pass filters respectively, and * denotes a convolution operation.

[0028] At each scale, the high-frequency subbands are decomposed in multiple directions, and the non-subsampled directional filter (NSDFB) is applied for directional filtering to capture the details and edge information of the image in different directions. By inverse filtering the subbands at each level, the detail subbands of each scale and direction are merged to reconstruct the high-frequency and low-frequency components. The reconstruction is performed step by step upwards, and finally the denoised image is restored.

[0029] Step S4, performing a line-by-line fast Fourier transform on the denoised image to obtain spectral data after denoising.

[0030] like Figure 8 As shown, the noise in the spectrum is effectively removed, and the denoising effect is evaluated by combining the peak signal-to-noise ratio and the root mean square error, as follows:

[0031] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to specific embodiments, a person skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention. The technologies, shapes, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. A spatial heterodyne interference correction method based on adaptive bilateral filtering and UDWT. The method is characterized in that: The following steps are involved: Step S1, preprocessing the original interferogram. Remove the influence of false low-frequency signals such as baseline, and remove some random noise that may affect spectral restoration; Step S2, adaptive bilateral filtering. Calculate the edge response of the interferogram image, introduce the result of edge detection into the convolution kernel weight, improve the convolution kernel so that it can dynamically adjust and enhance details, optimize the bilateral filtering weight parameters, and construct a bilateral filter. Convolve the preprocessed interferogram to achieve adaptive bilateral filtering; Step S3, two-dimensional non-subsampled wavelet transform (UD WT). Perform a two-dimensional non-subsampled wavelet transform on the interferogram after adaptive bilateral filtering to filter out the remaining noise; Step S4, spectral restoration. Take one row of the denoised spatial heterodyne interferogram for fast Fourier transform to obtain the denoised restored spectrum.

2. The adaptive bilateral filtering according to claim 1, characterized in that: When calculating the edge response in step S2, the Prewitt operator should be used to calculate the gradients of the horizontal and vertical directions of the image respectively. Then the gradients in these two directions are combined to determine the edge strength and direction of each pixel. The edge of the interference fringe indicates the area where the phase of the light changes dramatically. The Prewitt operator can capture the edge of the fringe well by detecting these change points.

3. The UDWT according to claim 1, characterized in that: In step S3, when performing wavelet transform, downsampling is not performed to avoid reduction in spatial resolution and ensure that the edges and features of the image will not be offset at different scales. The image should be decomposed into a low-frequency sub-band and three high-frequency sub-bands for each layer of decomposition, and the low-frequency sub-bands are iteratively decomposed. By inverse filtering the low-frequency and high-frequency sub-bands of each decomposition level, the sub-bands obtained by decomposition can be combined to reconstruct the denoised image.

Images