CN-122023193-A - Defogging and de-graining advanced method and device for optical remote sensing image

CN122023193ACN 122023193 ACN122023193 ACN 122023193ACN-122023193-A

Abstract

The invention provides a defogging and de-graining advanced method and device for an optical remote sensing image, aiming at the common problems of stripe noise and thin cloud fog shielding in the remote sensing image, providing an optimization treatment strategy of 'removing thin cloud fog firstly and removing stripes afterwards', and matching an improved algorithm, wherein on one hand, a segmentation self-adaptive window moment matching de-graining method based on an automatic sliding window is designed, an area is automatically divided through cloud detection, a gray segmentation threshold value and the sliding window size are dynamically determined, non-periodic stripes are effectively eliminated, on the other hand, a shallow U-Net structure de-cloud network with multi-scale cloud feature fusion is constructed, an on-chip color balance loss function is introduced, and image stitching chromatic aberration is inhibited. The invention can obviously improve the definition and visual consistency of the image, avoid the problem of cloud and fog residue caused by 'first stripping' or the problem of 'first stripping cloud and fog' and noise amplification, realize high-efficiency and high-quality remote sensing image restoration while maintaining texture details, and is suitable for the automatic processing of large-size high-resolution remote sensing data.

Inventors

QU XIAOFEI
WANG HONGGANG
LIN XIANGYANG
ZHANG FAN
ZHANG JIAXIN
WANG ANQI
Xiong Zhuolin

Assignees

北京市遥感信息研究所

Dates

Publication Date: 20260512
Application Date: 20260105

Claims (10)

1. The defogging and de-graining advanced method for the optical remote sensing image is characterized by comprising the following steps of: S1, performing thin cloud and fog removal processing on a remote sensing image by adopting a multi-scale cloud and fog feature fusion network, wherein the network comprises a multi-scale cloud and fog feature extraction module and a selective kernel fusion module, and reducing chromatic aberration at an image splicing position by minimizing a chromatic balance loss function; S2, carrying out segmentation processing on the image with thin cloud removed based on a cloud detection result, distinguishing cloud areas and cloud-free areas, respectively calculating column statistical parameters, and automatically determining a segmentation gray value according to the column statistical parameters; S3, determining the size of a sliding window by utilizing column mean absolute error curve analysis, setting dynamic window parameters for a particularly wide dark stripe region, and generating a stripe correction coefficient by weighting column mean and column variance calculation; And S4, performing stripe removal on the image by adopting a segmentation self-adaptive window moment matching algorithm, and respectively performing weighted gray value correction on the cloud area and the cloud-free area according to the stripe correction coefficient to eliminate the obvious stripe defect after defogging treatment.
2. The method of claim 1, wherein S1 comprises: S11, extracting features by a multi-scale cloud feature extraction module through convolution kernels of different scales, and carrying out feature enhancement by combining an enhanced parallel attention module; S12, the selective kernel fusion module fuses residual output and a network output result of the upper layer, so that capturing capability of different scale features and grasping capability of image details are improved.
3. The method of claim 1, wherein S2 comprises: s21, by reading the image with stripes and cloud mask data, distinguishing cloud areas and cloud-free areas pixel by pixel in a matching way; S22, performing row statistics on the cloud-free area, and taking the maximum value of the row statistics as a segmentation gray value DNseg.
4. The method of claim 1, wherein S3 comprises: S31, a window with 700 fixed columns is used when smoothing the average value of the image columns; S32, determining the position and width of the particularly wide dark stripe by searching the position and the wave trough size of the minimum wave trough in the absolute error curve, and setting the size of the sliding window to be 2 times of the wave trough size.
5. The method of claim 1, wherein S4 comprises: S41, according to the formula Calculating the gray value after stripe correction; and S42, respectively applying the formulas to the cloud area and the non-cloud area to perform weighted gray value correction.
6. The utility model provides a defogging removes line advanced device of optics remote sensing image which characterized in that includes: The system comprises a multi-scale cloud feature fusion module, a multi-scale cloud feature extraction module and a selective kernel fusion module, wherein the multi-scale cloud feature fusion module is used for carrying out thin cloud removal processing on a remote sensing image by adopting a multi-scale cloud feature fusion network, and the network comprises the multi-scale cloud feature extraction module and the selective kernel fusion module and reduces chromatic aberration at an image splicing part by minimizing a chromatic balance loss function; the cloud area segmentation processing module is used for carrying out segmentation processing on the image after thinning cloud fog based on a cloud detection result, distinguishing cloud areas and cloud-free areas, respectively calculating column statistical parameters, and automatically determining a segmentation gray value according to the column statistical parameters; the fringe correction parameter calculation module is used for determining the size of a sliding window by utilizing column mean absolute error curve analysis, setting dynamic window parameters for a particularly wide dark fringe area, and generating a fringe correction coefficient by weighting column mean and column variance calculation; The segmentation self-adaptive window moment matching processing module is used for removing stripes of the image by adopting a segmentation self-adaptive window moment matching algorithm, respectively executing weighted gray value correction on the cloud area and the cloud-free area according to the stripe correction coefficient, and eliminating the obvious stripe defect after defogging processing.
7. The apparatus of claim 6, wherein the multi-scale cloud feature fusion module is further to: extracting features through convolution kernels of different scales, and carrying out feature enhancement by combining with an enhanced parallel attention module; And the residual error output and the network output result of the upper layer are fused, so that the capturing capability of different scale features and the grasping capability of image details are improved.
8. The apparatus of claim 6, wherein the cloud zone segmentation processing module is further to: By reading the image with stripes and cloud mask data, distinguishing a cloud area and a cloud-free area pixel by pixel in a matching way; and carrying out row statistics on the cloud-free area, and taking the maximum value of the row statistics as a segmentation gray value DNseg.
9. A computer device comprising a processor and a memory; Wherein the processor executes a program corresponding to the executable program code by reading the executable program code stored in the memory, for implementing a defogging and striae-removing advanced method of an optical remote sensing image according to any one of claims 1-5.
10. A non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a defogging and de-graining method of an optical remote sensing image according to any one of claims 1 to 5.

Description

Defogging and de-graining advanced method and device for optical remote sensing image Technical Field The invention relates to the technical field of water image processing, in particular to a defogging and de-graining advanced method and device for an optical remote sensing image. Background The optical remote sensing image is used as an important component of the remote sensing technology and is widely applied to the fields of environment monitoring, disaster early warning, resource exploration and the like. In the related art, an optical remote sensing image preprocessing system is constructed through the cooperative operation of a defogging algorithm and a striping algorithm. Specifically, the technical system covers the whole process from image acquisition to quality optimization, and comprises key links such as cloud and fog detection, streak correction, color balance and the like. With the development of deep learning technology, a defogging method based on a neural network is gradually mainstream, but the defogging method faces a computational bottleneck when processing large-size images (the single data volume is usually more than 30G), and the conventional histogram matching and moment matching algorithms are difficult to cope with the problem of aperiodic stripes, so that a significant synergistic barrier exists in defogging and striping processing in the prior art. However, in the existing image processing method, the processing sequence of removing stripes and then removing fog is directly adopted, and the amplification effect of the defogging algorithm on noise is not fully considered, so that fine noise which is invisible to eyes in an original image can be fully learned and highlighted by a deep neural network in the defogging process. Specifically, the haze removal algorithm can amplify fine noise in an original image while removing haze, and the noise is superimposed with stripe defects after being processed, so that the subsequent stripping effect is remarkably reduced. The traditional moment matching algorithm can not effectively eliminate aperiodic stripes due to the fact that the space correlation of ground objects is not considered, and color difference stripes appear at the splicing position due to the fact that color balance constraint is lacking in large-image block processing. Due to the technical defects, the prior method is difficult to realize the cooperative optimization of defogging and stripping when processing complex ground object scenes such as water areas, towns and the like, and finally the consistency of image quality and the ground object recognition precision are affected. Disclosure of Invention The present invention aims to solve at least one of the technical problems in the related art to some extent. Therefore, a first objective of the present invention is to provide a defogging and de-graining method for an optical remote sensing image. Another objective of the present invention is to provide a defogging and de-graining step device for an optical remote sensing image. A third object of the invention is to propose a computer device. A fourth object of the present invention is to propose a non-transitory computer readable storage medium. In order to achieve the above objective, an embodiment of a first aspect of the present invention provides a defogging and striae removing method for an optical remote sensing image, including: S1, performing thin cloud and fog removal processing on a remote sensing image by adopting a multi-scale cloud and fog feature fusion network, wherein the network comprises a multi-scale cloud and fog feature extraction module and a selective kernel fusion module, and reducing chromatic aberration at an image splicing position by minimizing a chromatic balance loss function; S2, carrying out segmentation processing on the image with thin cloud removed based on a cloud detection result, distinguishing cloud areas and cloud-free areas, respectively calculating column statistical parameters, and automatically determining a segmentation gray value according to the column statistical parameters; S3, determining the size of a sliding window by utilizing column mean absolute error curve analysis, setting dynamic window parameters for a particularly wide dark stripe region, and generating a stripe correction coefficient by weighting column mean and column variance calculation; And S4, performing stripe removal on the image by adopting a segmentation self-adaptive window moment matching algorithm, and respectively performing weighted gray value correction on the cloud area and the cloud-free area according to the stripe correction coefficient to eliminate the obvious stripe defect after defogging treatment. In one embodiment of the present invention, the S1 includes: S11, extracting features by a multi-scale cloud feature extraction module through convolution kernels of different scales, and carrying out feature enhancement by combining an enhanced