CN109978855A - A kind of method for detecting change of remote sensing image and device - Google Patents

Abstract

Embodiments of the present invention provide a method and device for detecting changes in remote sensing images, the method comprising: obtaining a first remote sensing image and a second remote sensing image, the first remote sensing image and the second remote sensing image being shot at different times remote sensing images of the same area; generating a differential image according to the first remote sensing image and the second remote sensing image; removing noise in the differential image based on the frequency domain information of the differential image; The images are binarized to generate a change detection result, and the change detection result is used to indicate the difference between the first remote sensing image and the second remote sensing image. Since the information of the pixels in the differential image is rich in the frequency domain, the use of the frequency domain information of the differential image can remove the noise in the differential image, so that the differential image can truly represent the changes of the remote sensing image, improve the accuracy of the differential image classification, and then Improve the accuracy of change detection results.

Description

A kind of remote sensing image change detection method and device

technical field

The invention relates to the technical field of digital image processing, in particular to a method and device for detecting changes in remote sensing images.

Background technique

Remote sensing image change detection refers to the process of analyzing two remote sensing images obtained at different times in the same area to obtain the surface change characteristics of the area. It is one of the main development directions of current remote sensing data processing technology.

At present, the change detection of two remote sensing images obtained at different times in the same area is usually to generate a difference image (DI) first, and then perform binary classification on the difference image, that is, classify each pixel in the difference image. It is divided into change class or unchanged class, and then according to the division result of each pixel, the change detection results of two remote sensing images are obtained.

However, due to the existence of noise in the remote sensing image, there is also noise in the generated differential image, which makes the differential image unable to truly represent the changes of the remote sensing image, reduces the accuracy of the differential image classification, and leads to inaccurate change detection results. .

SUMMARY OF THE INVENTION

In view of the above problems, the purpose of the embodiments of the present invention is to provide a method and device for detecting changes in remote sensing images, aiming to remove noise in differential images, so that differential images can truly represent changes in remote sensing images, and improve the accuracy of differential image classification. improve the accuracy of change detection results.

In a first aspect, an embodiment of the present invention provides a method for detecting changes in a remote sensing image, the method comprising: obtaining a first remote sensing image and a second remote sensing image, the first remote sensing image and the second remote sensing image being at different times A remote sensing image of the same area taken; a differential image is generated according to the first remote sensing image and the second remote sensing image; based on the frequency domain information of the differential image, the noise in the differential image is removed; The difference images obtained are classified into two, and a change detection result is generated, and the change detection result is used to indicate the difference between the first remote sensing image and the second remote sensing image.

In a second aspect, an embodiment of the present invention provides a remote sensing image change detection device, the device includes: a receiving module configured to obtain a first remote sensing image and a second remote sensing image, the first remote sensing image and the second remote sensing image The remote sensing images are remote sensing images of the same area captured at different times; the image generation module is configured to generate a differential image according to the first remote sensing image and the second remote sensing image; the denoising module is configured to generate a differential image based on the The frequency domain information of the differential image, to remove noise in the differential image; the detection module is configured to perform binary classification on the differential image after noise removal, and generate a change detection result, where the change detection result is used to indicate the first The difference between the remote sensing image and the second remote sensing image.

In a third aspect, an embodiment of the present invention provides an electronic device, the electronic device includes: at least one processor; and at least one memory and a bus connected to the processor; wherein the processor and memory pass through the The bus completes mutual communication; the processor is configured to call program instructions in the memory to execute the method in one or more of the above technical solutions.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where the storage medium includes a stored program, wherein when the program runs, a device where the storage medium is located is controlled to execute one or more of the above technical solutions Methods.

In the method and device for detecting changes in remote sensing images provided by the embodiments of the present invention, after obtaining the first remote sensing image and the second remote sensing image, first, a differential image is generated according to the first remote sensing image and the second remote sensing image; then, a differential image is generated based on the differential image. The frequency domain information is used to remove the noise in the difference image; finally, the difference image after noise removal is classified into two categories to generate the change detection result. Since the information of the pixels in the differential image is rich in the frequency domain, the use of the frequency domain information of the differential image can remove the noise in the differential image, so that the differential image can truly represent the changes of the remote sensing image, improve the accuracy of the differential image classification, and then Improve the accuracy of change detection results.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for the purpose of illustrating preferred embodiments only and are not to be considered limiting of the invention. Also, the same components are denoted by the same reference numerals throughout the drawings. In the attached image:

1 is a schematic flowchart 1 of a method for detecting changes in remote sensing images in an embodiment of the present invention;

2 is a second schematic flowchart of a method for detecting changes in remote sensing images according to an embodiment of the present invention;

Fig. 3 is the remote sensing image of the Ottawa area in the embodiment of the present invention and the change detection diagram thereof;

Fig. 4 is the remote sensing image of Vietnam's Honghe region in the embodiment of the present invention and the change detection diagram thereof;

5 is a comparison diagram of the anti-noise performance of PCANet, NR-ELM, FDA-RMG, and NSST-APCNN relative to speckle noise in an embodiment of the present invention;

6 is a comparison diagram of the anti-noise performance of PCANet, NR-ELM, FDA-RMG, and NSST-APCNN relative to Gaussian noise in an embodiment of the present invention;

7 is a schematic structural diagram of a remote sensing image change detection device in an embodiment of the present invention;

FIG. 8 is a schematic structural diagram of an electronic device in an embodiment of the present invention.

Detailed ways

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present invention will be more thoroughly understood, and will fully convey the scope of the present invention to those skilled in the art.

Embodiments of the present invention provide a method and device for detecting changes in remote sensing images. In practical applications, when it is necessary to detect changes in the surface of a certain area at different times, remote sensing images of the area at two different times can be obtained first. Then, through the remote sensing image change detection method and device, the change detection results of the remote sensing images of the area at two different times are obtained, and then the surface changes of the area at the above two times are obtained.

Next, the method for detecting changes in remote sensing images provided by the embodiments of the present invention will be described in detail.

FIG. 1 is a schematic flowchart 1 of a method for detecting changes in remote sensing images according to an embodiment of the present invention. Referring to FIG. 1 , the method may include:

S101: Obtain a first remote sensing image and a second remote sensing image.

Here, the first remote sensing image and the second remote sensing image are remote sensing images of the same area captured at different times. For example, the first remote sensing image is the remote sensing image of area A taken on January 2, 2018, and the second remote sensing image is the remote sensing image of area A taken on March 19, 2019.

It should be noted here that the first remote sensing image and the second remote sensing image are images that have undergone geometric correction and geometric registration.

S102: Generate a difference image according to the first remote sensing image and the second remote sensing image.

In the specific implementation process, the gray values of all pixels in the first remote sensing image and the gray values of all pixels in the second remote sensing image may be obtained first, and then the gray values of all pixels in the first remote sensing image and the second remote sensing image are obtained. The gray value of all pixels in the remote sensing image is logarithmically calculated according to the spatial position correspondence to obtain a differential image. Here, the spatial position correspondence is the correspondence between the pixels in the first remote sensing image and the pixels in the second remote sensing image, wherein the position of a certain pixel in the first remote sensing image in the first remote sensing image and the second remote sensing image The position of a certain pixel in the image is the same in the second remote sensing image.

Exemplarily, it is assumed that there are only 4 pixels in both the first remote sensing image and the second remote sensing image, and the gray values of all pixels in the first remote sensing image are G ₁ , G ₂ , and G ₃ in order from left to right and from top to bottom. , G ₄ , the gray values of all pixels in the second remote sensing image are G ₅ , G ₆ , G ₇ , G ₈ from left to right and from top to bottom, then, the gray values of the obtained differential image are from left to Right from top to bottom

S103: Remove noise in the differential image based on the frequency domain information of the differential image.

Here, since it is not easy to identify noise from the image in the spatial domain, the differential image cannot be denoised very well, while in the frequency domain, the information of the pixels in the image is rich, which can be well extracted from the differential image. Identify the noise, and then remove the noise in the differential image, so that the differential image can truly represent the change of the remote sensing image, provide more accurate data for the second classification, and finally improve the accuracy of the change detection result.

S104: Perform two classifications on the difference image after noise removal to generate a change detection result.

Wherein, the change detection result is used to indicate the difference between the first remote sensing image and the second remote sensing image.

In a specific implementation process, a binary classification algorithm may be used to perform binary classification on the difference image after noise removal, and each pixel in the difference image after noise removal is classified into a change class or a non-change class. Here, the adopted binary classification algorithm is the prior art, and details are not repeated here. After each pixel in the difference image after noise removal is divided into a change class or an unchanged class, the pixel corresponding to the change class can be marked as 1, the pixel corresponding to the unchanged class can be marked as 0, or the change class can be marked as 0. The gray value of the corresponding pixel is set to 255, and the gray value of the pixel corresponding to the invariant class is set to 0. Then, the change detection result is a black and white image, and the white area in the image represents the first remote sensing image and the first remote sensing image. Two areas where changes occur between the remote sensing images, the black areas in the images represent the areas where there is no change between the first remote sensing image and the second remote sensing image. Of course, the pixels corresponding to the changed class and the pixels corresponding to the non-change class can also be distinguished in other ways, which are not limited here.

It can be seen from the above that, in the method for detecting changes in remote sensing images provided by the embodiments of the present invention, after obtaining the first remote sensing image and the second remote sensing image, first, a differential image is generated according to the first remote sensing image and the second remote sensing image; The frequency domain information of the difference image is used to remove the noise in the difference image; finally, the difference image after noise removal is classified into two categories to generate a change detection result. Since the information of the pixels in the differential image is rich in the frequency domain, the use of the frequency domain information of the differential image can remove the noise in the differential image, so that the differential image can truly represent the changes of the remote sensing image, improve the accuracy of the differential image classification, and then Improve the accuracy of change detection results.

Based on the foregoing embodiments, as a refinement and extension of the method shown in FIG. 1 , an embodiment of the present invention further provides a method for detecting changes in remote sensing images. FIG. 2 is a second schematic flowchart of a method for detecting changes in remote sensing images according to an embodiment of the present invention. Referring to FIG. 2 , the method may include:

S201: Obtain a first remote sensing image and a second remote sensing image.

S202: Perform denoising on the first remote sensing image and the second remote sensing image to obtain a denoised first remote sensing image and a denoised second remote sensing image.

In a specific implementation process, each pixel in the first remote sensing image may be processed through the neighborhood information of each pixel in the first remote sensing image, so as to achieve the purpose of denoising the first remote sensing image. Specifically, the neighborhood information of the point pixel (i, j) in the first remote sensing image whose size is H×W can be obtained through the following steps:

u=max(i-h,1)

d=min(i+h,H)

l=max(j-w,1)

r=min(j+w,W)

N=I(u:d,l:r)

X(i,j)=mean(N(:))

Among them, i∈[1,H], j∈[1,W], h, w are neighborhood size parameters, N is the neighborhood of the point pixel (i,j), X(i,j) is the point pixel ( i,j) neighborhood information.

Then, the information of the point pixel (i, j) can be re-determined according to the neighborhood information of the point pixel (i, j). Specifically, the mean or median value of the neighborhood information of the point pixel (i, j) can be calculated. As the information of the point pixel (i, j), of course, other processing can also be performed on the neighborhood information of the point pixel (i, j), and the processed result is used as the information of the point pixel (i, j). Do limit.

Here, since the neighborhood information of each pixel in the image is used to denoise the image, the interference of the noise on the differential image can be reduced, and the interference of the noise on the change detection result can be reduced, and the false alarm of the change detection result can be effectively reduced. quantity.

Similarly, the method for denoising the second remote sensing image is the same as the method for denoising the first remote sensing image, and details are not described herein again.

In practical applications, the mean filtering algorithm can be used to denoise the first remote sensing image and the second remote sensing image respectively; the median filtering algorithm can also be used to denoise the first remote sensing image and the second remote sensing image respectively; or both The mean filtering algorithm is used to denoise the first remote sensing image and the second remote sensing image respectively, and the median filtering algorithm is used to denoise the first remote sensing image and the second remote sensing image respectively, and then a filtering algorithm with a good denoising effect is selected. The first remote sensing image and the second remote sensing image are denoised respectively. Of course, other methods can also be used to denoise the first remote sensing image and the second remote sensing image respectively, which is not limited herein.

S203: Generate a difference image according to the denoised first remote sensing image and the denoised second remote sensing image.

S204: Perform frequency domain transformation on the difference image to obtain multiple sets of energy coefficients.

In the specific implementation process, a non-subsampled shearlet transform (NSST) can be used to transform the differential image into the frequency domain, because the non-subsampled shearlet transform has a scale when performing image rotation and translation transformation. Therefore, the non-subsampling shearlet transform can make the difference image keep all scales unchanged when it is transformed to the frequency domain, and maintain the accuracy of the frequency domain information of each scale of the differential image. The shearlet transform has low complexity and can save image processing time.

After transforming the difference image to the frequency domain, multiple energy coefficient sets can be obtained. The scales of the multiple energy coefficient sets here are different. The energy coefficient set of each scale can represent the difference image, but only for the difference image. The scales for representation are different, and each energy coefficient set includes multiple energy coefficients, and the directions of the multiple energy coefficients are different. Here, the energy coefficient in the frequency domain can be analogous to the gray value of the pixel in the space domain.

Exemplarily, assuming that the size of the difference image is 256×256, after transforming the difference image to the frequency domain, 128×128, 64×64, 32×32, 16×16, 8×8, 4×4 can be obtained , 2 × 2 energy coefficient sets of these scales, there are 128 × 128 energy coefficients in the energy coefficient set with a scale of 128 × 128, these 128 × 128 energy coefficients respectively represent the information of different positions in the differential image, which can be considered as the The difference image is divided into 128×128 small blocks, each small block has an energy coefficient, and the energy coefficient represents the information of the difference image at the position of the small block. The energy coefficient sets with scales of 64×64, 32×32, 16×16, 8×8, 4×4, and 2×2 are similar to the energy coefficient sets with scales of 128×128, and are not repeated here.

S205: Arrange all the energy coefficients in the multiple energy coefficient sets according to the absolute value, and set the energy coefficients smaller than the preset value to 0.

Since in the frequency domain, the energy coefficient of noise is small, by setting the energy coefficient with the smaller energy coefficient to 0, part of the noise in the differential image can be removed, so that the differential image can more realistically represent the first remote sensing The change between the image and the second remote sensing image, and then the difference image is classified into two, which can improve the accuracy of the change detection result.

Exemplarily, assuming that all the energy coefficients in the multiple energy coefficient sets are 4, 9, 8, 7, 6, 1, 3, 10, 2, and 5, these energy coefficients are arranged in descending order of absolute value as 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, because the points with small energy coefficients in the frequency domain are largely noise or points with small changes, therefore, the points less than 1.5 can be used. The energy coefficients are all set to 0, or you can set the energy coefficients of the bottom 10% to 0, resulting in 10, 9, 8, 7, 6, 5, 4, 3, 2, 0. The data in the above examples are only examples, and do not limit the embodiments of the present invention.

It should be noted here that the preset value can be set according to the actual detection requirements. If a large change needs to be detected, the preset value can be set larger. If a small change needs to be detected, the preset value can be set. The value is set smaller. In practical applications, the energy coefficient of the last 10% in the row from largest to smallest is generally set to 0. In this way, global noise reduction can be performed, and as many points that change in the differential image can be retained as much as possible. .

S206: Perform noise estimation on each energy coefficient set to obtain the noise variance of each energy coefficient set.

Here, the energy coefficients larger than the preset value in each energy coefficient set remain unchanged, and the energy coefficients smaller than the preset value have been set to 0.

S207: Based on the noise variance, perform denoising on the energy coefficient set corresponding to the noise variance.

In the specific implementation process, noise estimation can be performed on each energy coefficient set based on the last stage of wavelet transform to obtain the noise variance σ of each energy coefficient set, where the noise variance σ can represent the corresponding energy coefficient Concentrated noise intensity. Since the noise variance of the energy coefficient sets of different scales is different, according to the noise variance of the energy coefficient set of each scale, denoising the energy coefficient set of the scale corresponding to the noise variance can retrieve the mistakenly regarded as noise and delete it. To avoid the one-size-fits-all situation, that is, to avoid using the same noise variance to delete the truly changed points as noise, it can effectively reduce the number of missed alarms in the change detection results, thereby improving the accuracy of the change detection results.

Exemplarily, it is assumed that the noise variance in the energy coefficient set A is 2, and the noise variance in the energy coefficient set B is 0.2. If a "one size fits all" global denoising method is adopted, that is, the global denoising threshold is set to 2, then the denoising effect for the energy coefficient set A is very good, but the denoising effect for the energy coefficient set B is very poor. Because the "one-size-fits-all" denoising method removes the points that are originally signals in the energy coefficient set B and mistakenly regarded as noise, which leads to an increase in the false detection rate. If the local denoising method of "differential treatment" is adopted, that is, the denoising threshold of energy coefficient set A is set to 2, and the denoising threshold of energy coefficient set B is set to 0.2. In this way, not only the denoising effect of energy coefficient set A is very good , and the denoising effect of the energy coefficient set B is also very good, so that the denoising effect of the entire energy coefficient set is very good.

S208: Perform inverse frequency domain transformation on the multiple energy coefficient sets after denoising, to obtain a differential image after inverse frequency domain transformation.

In a specific implementation process, the non-subsampling inverse shear wave transform can be used to transform the multiple energy coefficient sets after denoising into the spatial domain, so as to obtain a differential image after inverse transformation in the frequency domain. The non-subsampling inverse shearlet transform is the inverse process of the non-subsampling shearlet transform. The non-subsampling inverse shearlet transform also has the characteristics of scale invariance when performing image rotation and translation transformation. The inverse shearlet transform can keep the scales unchanged when the differential image is changed back to the spatial domain, and maintain the accuracy of the spatial information of the differential image at each scale. Image processing time.

It should be noted here that, for the frequency domain transformation performed in step S204 and the frequency domain inverse transformation performed in step S208, the non-downsampling shear wave transformation in step S204 and the non-downsampling in step S208 may be adopted. The inverse shear wave transform can also be the non-subsampling shear wave transform in step S204 and other frequency domain inverse transforms except the non-downsampling inverse shear wave transform in step S208, or the step S204 In step S208, other frequency domain transforms other than the non-downsampling shearlet transform are used, and the non-downsampling shearlet inverse transform is used in step S208, and other frequency domain transforms other than the non-downsampling shearlet transform can also be used in step S204. The transformation and other frequency domain inverse transforms other than the down-sampled shear wave inverse transform are used in step S208, which are not limited here.

S209: Perform binary classification on the difference image after inverse frequency domain transformation to generate a change detection result.

In a specific implementation process, an adaptive pulse coupled neural network (Adaptive PulseCoupled Neural Network, APCNN) may be used to perform binary classification on the difference image after noise removal to generate a change detection result. Since the adaptive pulse coupled neural network can automatically determine the threshold value according to the characteristics of the image when classifying the image, using the adaptive pulse coupled neural network to perform binary classification on the difference image after noise removal can improve the performance of the difference image. At the same time, the adaptive pulse coupled neural network also has the function of denoising, and it can also denoise the difference image again, improve the anti-noise performance of the method, and improve the accuracy of the binary classification of the difference image. In addition, The low complexity of the adaptive impulse coupled neural network can also save image processing time.

So far, according to the generated change detection results, it is possible to know the difference between the first remote sensing image and the second remote sensing image, and then to know the surface changes in a certain area.

The detection effect of the remote sensing image change detection method is described below with a specific example.

Example 1: Figure 3 is a remote sensing image of the Ottawa area and its change detection map in the embodiment of the present invention. Referring to Figure 3, 3a is a remote sensing image of the Ottawa area obtained in May 1997, and 3b is a remote sensing image of the Ottawa area obtained in May 1997. The remote sensing images of the Ottawa area obtained in 1997, 3c is the change detection result obtained from May 1997 to August 1997 according to the actual changes of the surface of Ottawa, that is, the reference image, 3d is based on Gao Feng and other scholars in the article "Automatic The PCANet-based change detection method proposed in Change Detectionin Synthetic Aperture Radar Images Based on PCANet, referred to as PCANet, and the obtained change detection results between 3a and 3b, 3e is based on Gao Feng and other scholars in the article "Change detection from synthetic The change detection algorithm based on neighborhood logarithm ratio and extreme learning machine proposed in aperture radar images based onneighborhood-based ratio and extreme learning machine, referred to as NR-ELM, and the obtained change detection result between 3a and 3b, 3f According to the change detection algorithm based on frequency-domain analysis and random multigraphs proposed by Gao Feng and other scholars in the article "Synthetic aperture radar image change detection based on frequency-domain analysis and random multigraphs", referred to as FDA-RMG, the obtained 3a and The change detection result between 3b and 3g is the change detection result between 3a and 3b obtained by the remote sensing image change detection method provided according to the embodiment of the present invention, referred to as NSST-APCNN for short.

In terms of visual effects, 3d, 3e, 3f, 3g and 3c are compared. The white area represents the changed area, and the black area represents the unchanged area. By comparison, the detection result of 3g is slightly better than that of 3d and 3f. It is much better than 3e, that is to say, by using the remote sensing image change detection method provided by the embodiment of the present invention, the change between two remote sensing images can be detected more accurately.

In terms of objective indicators, compare 3d, 3e, 3f, and 3g, see Table 1. Table 1 is the objective indicators of 3d, 3e, 3f, and 3g. Among them, the objective indicators of 3d, 3e, 3f, and 3g are all Obtained on the basis of 3c.

Table 1

Among them, the Kappa coefficient is used for the consistency test, the recall rate is the recall rate, and F1 is an indicator used to measure the accuracy of the two-class model in statistics, taking into account the accuracy and recall rate of the classification model, which can be regarded as is a weighted average of model precision and recall.

By comparing 3g with 3d, 3e, and 3f, it can be seen that although the number of false alarm pixels in 3g has increased compared with that in 3e and 3f, the number of missed pixels in 3g and the total number of false pixels in 3g are higher than those in 3d, 3e, and 3f. Compared with 3d, 3e, and 3f, the correct classification percentage, Kappa coefficient, recall rate, and F1 of 3g have increased, and the running time of 3g has been greatly reduced compared with 3d, 3e, and 3f, that is to say, Compared with 3d, 3e, and 3f, 3g has improved accuracy of change detection results and reduced running time.

It can be seen that, whether in terms of visual effects or objective indicators, using the remote sensing image change detection method provided by the embodiments of the present invention can improve the accuracy of remote sensing image change detection results and reduce the running time.

Example two: Fig. 4 is the remote sensing image of the Vietnam Honghe area in the embodiment of the present invention and the change detection diagram thereof, referring to shown in Fig. 4, 4a is the remote sensing image of the Vietnam Honghe area obtained on August 24, 1996, 4b is The remote sensing image of the Red River area of Vietnam obtained on August 14, 1999, 4c is the change detection result obtained from August 24, 1996 to August 14, 1999 according to the actual change of the surface of the Red River in Vietnam, that is, the reference map , 4d is the PCANet-based change detection method proposed by Gao Feng and other scholars in the article "Automatic Change Detection in Synthetic Aperture Radar Images Based on PCANet", referred to as PCANet, and the obtained change detection results between 4a and 4b, 4e is According to the change detection algorithm based on the neighborhood logarithm ratio and extreme learning machine (NR-ELM for short) proposed by Gao Feng and other scholars in the article "Change detection from synthetic apertureradar images based on neighborhood-based ratio and extreme learning machine", we get The change detection results between 4a and 4b, 4f is the change detection based on frequency-domain analysis and random multigraphs proposed by Gao Feng et al. in the article "Synthetic aperture radar imagechange detection based on frequency-domain analysis and random multigraphs" Algorithm, FDA-RMG for short, and the obtained change detection result between 4a and 4b, 4g is the remote sensing image change detection method provided according to the embodiment of the present invention, referred to as NSST-APCNN, and the obtained change between 4a and 4b Test results.

In terms of visual effects, 4d, 4e, 4f, 4g and 4c are compared. The white area represents the changed area, and the black area represents the unchanged area. By comparison, the detection result of 4g is slightly better than that of 4e and 4f. It is much better than 4d, that is to say, by using the remote sensing image change detection method provided by the embodiment of the present invention, the change between two remote sensing images can be detected more accurately.

In terms of objective indicators, compare 4d, 4e, 4f, and 4g, see Table 2. Table 2 is the objective indicators of 4d, 4e, 4f, and 4g. Among them, the objective indicators of 4d, 4e, 4f, and 4g are all Obtained on the basis of 4c.

Table 2

Among them, the Kappa coefficient is used for the consistency test, the recall rate is the recall rate, and F1 is an indicator used to measure the accuracy of the two-class model in statistics, taking into account the accuracy and recall rate of the classification model, which can be regarded as is a weighted average of model precision and recall.

By comparing 4g with 4d, 4e, and 4f, it can be seen that although the number of false alarm pixels in 4g has increased compared with that in 4d, 4e, and 4f, the number of missed pixels and total error pixels in 4g are higher than those in 4d and 4e. , 4f are reduced, the correct classification percentage, Kappa coefficient, recall rate, and F1 of 4g have increased compared with 4d, 4e, and 4f, and the running time of 4g has been greatly reduced compared with 4d, 4e, and 4f, that is, Said, 4g compared to 4d, 4e, 4f, the accuracy of change detection results has been improved, and the running time has been reduced.

It can be seen that, whether in terms of visual effects or objective indicators, using the remote sensing image change detection method provided by the embodiments of the present invention can improve the accuracy of remote sensing image change detection results and reduce the running time.

Next, the anti-noise performance of the remote sensing image change detection method is further explained. Here, the anti-noise performance refers to the degree to which the transformation detection result changes before and after adding noise to the remote sensing image. The smaller the degree of change, the closer to 1. The better the anti-noise performance.

FIG. 5 is a comparison diagram of the anti-noise performance of PCANet, NR-ELM, FDA-RMG, and NSST-APCNN relative to speckle noise in an embodiment of the present invention. Referring to FIG. 5, 5a is a speckle added to the remote sensing image in the Ottawa area. The comparison chart of anti-noise performance obtained from noise. 5b is the comparison chart of anti-noise performance obtained by adding speckle noise to the remote sensing image of the Honghe region of Vietnam. The range of the added speckle noise is PSNR∈[26,51]dB, and PSNR is the peak signal-to-noise ratio. , namely Peak Signal to NoiseRatio, τ is the anti-noise performance, whether in 5a or 5b, it can be seen that the anti-noise performance of NSST-APCNN is slightly higher than that of PCANet and FDA-RMG, and higher than that of NR-ELM.

FIG. 6 is a comparison diagram of the anti-noise performance of PCANet, NR-ELM, FDA-RMG, and NSST-APCNN relative to Gaussian noise in the embodiment of the present invention. Referring to FIG. 6, 6a is the addition of Gaussian to the remote sensing image in the Ottawa area. The comparison chart of anti-noise performance obtained from noise. 6b is the comparison chart of anti-noise performance obtained by adding Gaussian noise to the remote sensing image of the Honghe region of Vietnam. The range of the added Gaussian noise is PSNR∈[35,50]dB, and PSNR is the peak signal-to-noise ratio. , namely Peak Signal to NoiseRatio, τ is the anti-noise performance, whether in 6a or 6b, it can be seen that the anti-noise performance of NSST-APCNN is slightly higher than that of PCANet and FDA-RMG, and higher than that of NR-ELM.

It can be seen from the above that whether it is speckle noise or Gaussian noise, in the range of PSNR ∈ [35, 50] dB, the anti-noise performance of NSST-APCNN is slightly higher than that of PCANet and FDA-RMG, and higher than that of NR-ELM.

Based on the same inventive concept, as an implementation of the above method, an embodiment of the present invention further provides a remote sensing image change detection device. FIG. 7 is a schematic structural diagram of a remote sensing image change detection device in an embodiment of the present invention. Referring to FIG. 7 , the device 70 may include: a receiving module 701, configured to obtain a first remote sensing image and a second remote sensing image, the first remote sensing image The remote sensing image and the second remote sensing image are remote sensing images of the same area captured at different times; the image generation module 702 is configured to generate a differential image according to the first remote sensing image and the second remote sensing image; the denoising module 703 is configured as Based on the frequency domain information of the differential image, the noise in the differential image is removed; the detection module 704 is configured to perform binary classification on the differential image after noise removal, and generate a change detection result, and the change detection result is used to indicate the difference between the first remote sensing image and the first remote sensing image. 2. Differences between remote sensing images.

Based on the foregoing embodiment, the denoising module is configured to perform frequency domain transformation on the differential image to obtain a plurality of energy coefficient sets, the scales of the plurality of energy coefficient sets are different, and each energy coefficient set includes a plurality of energy coefficients. The directions of the energy coefficients are different; all the energy coefficients in the multiple energy coefficient sets are arranged according to the absolute value, and the energy coefficients smaller than the preset value are set to 0; the energy coefficients larger than the preset value are set Perform inverse frequency domain transformation with an energy coefficient of 0 to obtain an inversely transformed differential image in the frequency domain; the detection module is configured to perform binary classification on the differential image after inverse transformation in the frequency domain to generate a change detection result.

Based on the foregoing embodiment, the denoising module is further configured to perform noise estimation on each energy coefficient set to obtain the noise variance of each energy coefficient set; based on the noise variance, perform denoising on the energy coefficient set corresponding to the noise variance; The multiple energy coefficient sets after denoising are inversely transformed in the frequency domain to obtain a differential image after the inverse frequency domain transformation.

Based on the foregoing embodiment, the denoising module is configured to perform frequency domain transformation on the differential image based on the non-subsampled shearlet transform to obtain multiple sets of energy coefficients.

Based on the foregoing embodiment, the denoising module is configured to perform inverse frequency domain transformation on energy coefficients greater than a preset value and energy coefficients set to 0 based on non-subsampled shear wave inverse transform, and obtain the frequency domain inverse transform difference image.

Based on the foregoing embodiment, the detection module is configured to perform binary classification on the difference image after noise removal based on the adaptive pulse coupled neural network, and generate a change detection result.

Based on the foregoing embodiment, the apparatus may further include: a preprocessing module; a preprocessing module configured to perform denoising on the first remote sensing image and the second remote sensing image to obtain a denoised first remote sensing image and a denoised remote sensing image. a second remote sensing image; an image generating module configured to generate a differential image according to the denoised first remote sensing image and the denoised second remote sensing image.

It should be pointed out here that the descriptions of the above apparatus embodiments are similar to the descriptions of the above method embodiments, and have similar beneficial effects to the method embodiments. For technical details not disclosed in the apparatus embodiments of the present invention, please refer to the description of the method embodiments of the present invention to understand.

Based on the same inventive concept, an embodiment of the present invention also provides an electronic device. FIG. 8 is a schematic structural diagram of an electronic device in an embodiment of the present invention. Referring to FIG. 8, the electronic device 80 may include: at least one processor 801; and at least one memory 802 and a bus 803 connected to the processor 801; wherein , the processor 801 and the memory 802 communicate with each other through the bus 803; the processor 801 is configured to call program instructions in the memory 802 to execute the method in one or more of the above embodiments.

It should be pointed out here that the descriptions of the above electronic device embodiments are similar to the descriptions of the above method embodiments, and have similar beneficial effects to the method embodiments. For technical details not disclosed in the embodiments of the electronic device of the embodiments of the present invention, please refer to the description of the method embodiments of the present invention for understanding.

Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored program, wherein, when the program runs, the device where the storage medium is located is controlled to execute one or more of the foregoing embodiments method in .

It should be pointed out here that the descriptions of the above computer-readable storage medium embodiments are similar to the descriptions of the above method embodiments, and have similar beneficial effects to the method embodiments. For technical details not disclosed in the embodiments of the computer-readable storage medium of the embodiments of the present invention, please refer to the description of the method embodiments of the present invention for understanding.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

Memory may include non-persistent memory in computer readable media, random access memory (RAM) and/or non-volatile memory in the form of, for example, read only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media includes both persistent and non-permanent, removable and non-removable media, and storage of information may be implemented by any method or technology. Information may be computer readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Flash Memory or other memory technology, Compact Disc Read Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

It should also be noted that the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device comprising a series of elements includes not only those elements, but also Other elements not expressly listed, or which are inherent to such a process, method, article of manufacture, or apparatus are also included. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article of manufacture or apparatus that includes the element.

It will be appreciated by those skilled in the art that the embodiments of the present application may be provided as a method, a system or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The above are merely examples of the present application, and are not intended to limit the present application. Various modifications and variations of this application are possible for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the scope of the claims of this application.

Claims (10)

1. a remote sensing image change detection method, is characterized in that, described method comprises: obtaining a first remote sensing image and a second remote sensing image, where the first remote sensing image and the second remote sensing image are remote sensing images of the same area captured at different times; generating a differential image according to the first remote sensing image and the second remote sensing image; removing noise in the differential image based on the frequency domain information of the differential image; Binary classification is performed on the difference image after noise removal, and a change detection result is generated, and the change detection result is used to indicate the difference between the first remote sensing image and the second remote sensing image. 2. The method according to claim 1, wherein the removing noise in the differential image based on the frequency domain information of the differential image comprises: Perform frequency domain transformation on the differential image to obtain a plurality of energy coefficient sets, the scales of the plurality of energy coefficient sets are different, each energy coefficient set includes a plurality of energy coefficients, and the directions of the plurality of energy coefficients are different. Are not the same; Arranging all the energy coefficients in the plurality of energy coefficient sets according to the absolute value, and setting the energy coefficients smaller than the preset value to 0; Perform inverse frequency domain transformation on the energy coefficient greater than the preset value and the energy coefficient set to 0 to obtain a differential image after inverse frequency domain transformation; The second classification is performed on the difference image after noise removal to generate a change detection result, including: Binary classification is performed on the difference image after inverse frequency domain transformation to generate a change detection result. 3 . The method according to claim 2 , wherein the inverse frequency domain transform is performed on the energy coefficient greater than the preset value and the energy coefficient set to 0 to obtain a differential image after the inverse frequency domain transformation. 4 . Before, the method further includes: Perform noise estimation on each energy coefficient set to obtain the noise variance of each energy coefficient set; Based on the noise variance, denoising the energy coefficient set corresponding to the noise variance; The performing inverse frequency domain transformation on the energy coefficient greater than the preset value and the energy coefficient set to 0 to obtain an inversely transformed differential image in the frequency domain, including: Perform inverse frequency domain transformation on the multiple energy coefficient sets after denoising, and obtain the differential image after inverse frequency domain transformation. 4. The method according to claim 2 or 3, wherein the performing frequency domain transformation on the differential image to obtain a plurality of energy coefficient sets, comprising: Based on the non-subsampled shearlet transform, frequency domain transform is performed on the differential image to obtain multiple sets of energy coefficients. 5 . The method according to claim 4 , wherein the inverse frequency domain transform is performed on the energy coefficient greater than the preset value and the energy coefficient set to 0 to obtain a differential image after the inverse frequency domain transformation. 6 . ,include: Based on the non-subsampling shear wave inverse transform, the frequency domain inverse transform is performed on the energy coefficients greater than the preset value and the energy coefficients set to 0 to obtain the frequency domain inversely transformed differential image. 6 . The method according to claim 2 or 3 , wherein the inverse frequency domain transform is performed on the energy coefficient greater than the preset value and the energy coefficient set to 0 to obtain the frequency domain inverse transform. 7 . Difference image, including: Based on the non-subsampling shear wave inverse transform, the frequency domain inverse transform is performed on the energy coefficients greater than the preset value and the energy coefficients set to 0 to obtain the frequency domain inversely transformed differential image. 7. The method according to any one of claims 1 to 3, wherein the performing binary classification on the noise-removed differential image to generate a change detection result, comprising: Based on the adaptive pulse coupled neural network, the difference image after denoising is binarized to generate the change detection result. 8. The method according to any one of claims 1 to 3, wherein, before generating a differential image according to the first remote sensing image and the second remote sensing image, the method further comprises: Denoising the first remote sensing image and the second remote sensing image to obtain a denoised first remote sensing image and a denoised second remote sensing image; The generating a differential image according to the first remote sensing image and the second remote sensing image, comprising: A differential image is generated according to the denoised first remote sensing image and the denoised second remote sensing image. 9. A remote sensing image change detection device, wherein the device comprises: a receiving module, configured to obtain a first remote sensing image and a second remote sensing image, wherein the first remote sensing image and the second remote sensing image are remote sensing images of the same area captured at different times; an image generation module configured to generate a differential image according to the first remote sensing image and the second remote sensing image; a denoising module configured to remove noise in the differential image based on the frequency domain information of the differential image; The detection module is configured to perform binary classification on the difference image after noise removal, and generate a change detection result, where the change detection result is used to indicate the difference between the first remote sensing image and the second remote sensing image. 10. An electronic device, characterized in that the electronic device comprises: at least one processor; and at least one memory and bus connected to the processor; Wherein, the processor and the memory communicate with each other through the bus; the processor is configured to call program instructions in the memory to execute the method according to any one of claims 1 to 6.