Soil pollution plot monitoring method based on remote sensing information, terminal and medium

Abstract

The application provides a soil pollution plot monitoring method based on remote sensing information, a terminal and a medium, wherein the soil pollution plot monitoring method comprises the following steps: obtaining remote sensing data corresponding to a current time and a reference time of a target plot respectively, and performing preprocessing, and performing color feature separation to obtain hue channel data, saturation channel data and lightness channel data corresponding to the current time and the reference time respectively; performing modal decomposition operation based on the saturation channel data and the lightness channel data corresponding to the current time and the reference time respectively, obtaining a plurality of current effective feature components and a plurality of reference effective feature components, and combining the corresponding hue channel data to perform feature fusion, and obtaining a current enhanced feature map and a reference enhanced feature map; performing change detection based on the current enhanced feature map and the reference enhanced feature map, and performing labeling to obtain the pollution change of the target plot, so as to realize accurate and effective plot pollution monitoring.

Description

Technical Field

[0001] This application belongs to the field of remote sensing image monitoring, and relates to a soil contaminated site monitoring technology, and in particular to a soil contaminated site monitoring method, terminal and medium based on remote sensing information. Background Technology

[0002] Land monitoring based on remote sensing information can achieve comprehensive, timely, and high-precision monitoring. In particular, for soil-contaminated sites, land monitoring enables systematic investigation, assessment, and long-term supervision, which can help to remediate and redevelop soil-contaminated sites.

[0003] Currently, monitoring methods for contaminated sites typically rely on remote sensing images acquired at the current and baseline times to detect changes and assess the site's state relative to the baseline. This allows for effective monitoring. However, it's important to note that remote sensing images acquired at different times exhibit significant variations due to environmental factors such as weather and lighting conditions. This means that pseudo-changes exist between images acquired at different times, and these pseudo-changes typically need to be removed to prevent them from affecting change detection and thus inaccurate results, ultimately impacting the monitoring effectiveness. Currently, however, there is no optimal method for removing image pseudo-changes.

[0004] Therefore, how to quickly and efficiently identify changes in soil-contaminated sites using satellite imagery data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide a method, terminal and medium for monitoring soil contaminated sites based on remote sensing information, in order to solve the problem that the detection accuracy of remote sensing image change detection in the prior art is not high, resulting in poor monitoring effect of contaminated sites.

[0006] Firstly, this application provides a method for monitoring soil contaminated sites based on remote sensing information, including:

[0007] The remote sensing data of the target plot at the current time and the reference time are acquired and preprocessed respectively;

[0008] Color feature separation is performed based on the preprocessed remote sensing data to obtain the current hue channel data, current saturation channel data, and current brightness channel data corresponding to the current time, as well as the reference hue channel data, reference saturation channel data, and reference brightness channel data corresponding to the reference time.

[0009] Modal decomposition operations are performed based on the current saturation channel data and the current brightness channel data to obtain several current effective feature components; and modal decomposition operations are performed based on the reference saturation channel data and the reference brightness channel data to obtain several reference effective feature components.

[0010] Based on each of the current effective feature components and combined with the current tone channel data, feature fusion is performed to obtain the current enhanced feature map; based on each of the benchmark effective feature components and combined with the benchmark tone channel data, feature fusion is performed to obtain the benchmark enhanced feature map.

[0011] Change detection is performed based on the current enhanced feature map and the baseline enhanced feature map, and the changes are marked to obtain the pollution changes of the target plot.

[0012] In one embodiment of this application, for any one of the current saturation channel data, the current brightness channel data, the reference saturation channel data, and the reference brightness channel data, the modal decomposition operation includes:

[0013] Two-dimensional empirical mode decomposition is performed on the data to be processed to obtain several mid-frequency intrinsic mode function components, several low-frequency intrinsic mode function components, and global residual components; wherein, the data to be processed is any one of the current saturation channel data, the current brightness channel data, the reference saturation channel data, and the reference brightness channel data;

[0014] Based on the two-dimensional local mean decomposition of each of the mid-frequency intrinsic mode function components, several product function components and local residuals corresponding to each of the mid-frequency intrinsic mode function components are obtained;

[0015] Each of the product function components, each of the local residuals, each of the low-frequency intrinsic mode function components, and the global residual components are used as each of the effective feature components; wherein, when the data to be processed is the current saturation channel data or the current brightness channel data, each of the product function components, each of the local residuals, each of the low-frequency intrinsic mode function components, and the global residual components are all used as each of the current effective feature components; when the data to be processed is the reference saturation channel data or the reference brightness channel data, each of the product function components, each of the local residuals, each of the low-frequency intrinsic mode function components, and the global residual components are all used as each of the reference effective feature components.

[0016] In one embodiment of this application, the step of performing two-dimensional empirical mode decomposition on the data to be processed to obtain several mid-frequency intrinsic mode function components, several low-frequency intrinsic mode function components, and global residual components is a cyclic decomposition process. A single decomposition process includes:

[0017] Obtain the current data to be decomposed and the current number of decompositions, and combine them with the preset local region size to obtain each local maximum point and each local minimum point;

[0018] Based on each of the local maxima, the upper envelope is fitted and obtained; and based on each of the local minima, the lower envelope is fitted and obtained.

[0019] Based on the upper and lower envelope surfaces, a local mean surface is obtained. The difference between the current data to be decomposed and the local mean surface is obtained as the current first residual data. The standard deviation of the current first residual data and the current data to be decomposed is calculated. If the standard deviation is within the stopping interval, the current first residual data is output as an intrinsic mode function, the current decomposition count is output as the corresponding decomposition count, the difference between the data to be processed and the current first residual data is output as the global residual component, and the decomposition stops. Otherwise, the current first residual data is output as the intrinsic mode function component, the current decomposition count is output as the corresponding decomposition count, the decomposition count is incremented by one as the new current decomposition count, the intrinsic mode function component is used as the new current data to be decomposed, and the next decomposition process is executed again.

[0020] Specifically, for each intrinsic mode function component, the corresponding dominant wavelength is obtained through Fourier transform. If the number of decompositions corresponding to any intrinsic mode function component exceeds a preset threshold and the corresponding dominant wavelength is within the mid-frequency wavelength range, then the intrinsic mode function component is a mid-frequency intrinsic mode function component. If the number of decompositions corresponding to any intrinsic mode function component exceeds a preset threshold and the corresponding dominant wavelength is within the low-frequency wavelength range, then the intrinsic mode function component is a low-frequency intrinsic mode function component.

[0021] In one embodiment of this application, for any of the mid-frequency intrinsic mode function components, a two-dimensional local mean decomposition is performed, which is a cyclic decomposition process. In a single decomposition process, the following is included:

[0022] Based on the current intrinsic mode function components to be decomposed, extract the corresponding product function components and output them;

[0023] The current local residual component is obtained based on the difference between the current eigenmode function component to be decomposed and the product function component. If the current local residual component satisfies the monotonicity condition, the current local residual component is output as the local residual component. Otherwise, the current local residual component is used as the new current eigenmode function component to be decomposed, and the next decomposition process is executed again.

[0024] In one embodiment of this application, the extraction of the product function components is a cyclic extraction process, and a single extraction process includes:

[0025] The system acquires the current instantaneous amplitude and the current demodulated signal; based on the current demodulated signal and a preset local region size for each component, it acquires the local maxima and local minima for each component; based on the local maxima, it fits and acquires the upper envelope of the component; based on the local minima, it fits and acquires the lower envelope of the component; based on the upper and lower envelopes, it acquires the corresponding local mean and local envelope of the component.

[0026] The difference between the current demodulated signal and the component local mean surface is obtained, and normalized based on the component local envelope surface to obtain a new demodulated signal. If the component local envelope surface is constant at one, the product of the component local envelope surface, the current instantaneous amplitude, and the new demodulated signal is used as the product function component output, and the extraction process stops. Otherwise, the product of the component local envelope surface and the current instantaneous amplitude is used as the new current instantaneous amplitude, and the new demodulated signal is used as the new current demodulated signal, and the next extraction process is executed again.

[0027] In one embodiment of this application, the step of performing feature fusion based on each of the currently effective feature components and the current tone channel data to obtain the current enhanced feature map includes:

[0028] The current multi-channel feature map is obtained by fusing the current effective feature components corresponding to the current saturation channel data, the current effective feature components corresponding to the current brightness channel data, and the current hue channel data respectively.

[0029] Calculate the energy value and gradient value of each of the current effective feature components to obtain the weight corresponding to each of the current effective feature components;

[0030] Based on the weights corresponding to each of the current effective feature components, the multi-channel feature map is convolved and weighted to obtain the current enhanced feature map.

[0031] The step of performing feature fusion based on each of the aforementioned effective feature components and combining the aforementioned hue channel data to obtain a reference enhanced feature map includes:

[0032] The reference multi-channel feature map is obtained by fusing the reference effective feature components corresponding to the reference saturation channel data, the reference effective feature components corresponding to the reference brightness channel data, and the reference hue channel data respectively.

[0033] Calculate the energy value and gradient value of each of the aforementioned effective feature components to obtain the weights corresponding to each of the aforementioned effective feature components;

[0034] Based on the weights corresponding to each of the aforementioned effective feature components, the multi-channel feature map is convolved and weighted to obtain the benchmark enhanced feature map.

[0035] In one embodiment of this application, the step of performing change detection and annotation based on the current enhanced feature map and the reference enhanced feature map to obtain the pollution change status of the target plot includes:

[0036] Based on a pre-trained change detection network, the current enhanced feature map and the baseline enhanced feature map are input to divide each region and obtain the corresponding annotation information for each region;

[0037] The current enhanced feature map is restored to obtain an RGB image, and the RGB image is annotated in combination with the annotation information corresponding to each region, so as to output an annotated RGB image reflecting the pollution change of the target plot.

[0038] In one embodiment of this application, after obtaining the current tone channel data and the reference tone channel data, the method further includes performing smoothing and noise reduction processing on both the current tone channel data and the reference tone channel data.

[0039] Secondly, this application provides a terminal, including: a processor and a memory, wherein the memory and the processor are communicatively connected;

[0040] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal performs the soil pollution site monitoring method based on remote sensing information as described above.

[0041] Thirdly, this application provides a computer storage medium storing a computer program, which, when executed by a processor, implements the aforementioned method for monitoring soil contaminated sites based on remote sensing information.

[0042] As described above, this application provides a method, terminal, and medium for monitoring soil contaminated sites based on remote sensing information. By separating color features, saturation channel data and brightness channel data are extracted and subjected to modal decomposition to obtain each effective feature component. This component is then fused with the hue channel data to form an enhanced feature map. This not only effectively removes false changes in the image but also enhances the features, thereby improving the accuracy of change detection and enhancing the monitoring effect of contaminated sites. Attached Figure Description

[0043] Figure 1 The diagram shown is a flowchart illustrating a method for monitoring soil contaminated sites as described in an embodiment of this application.

[0044] Figure 2 The diagram shown is a flowchart illustrating a mode decomposition operation as described in an embodiment of this application.

[0045] Figure 3 The diagram shown is a flowchart illustrating a single decomposition process in the execution of a two-dimensional empirical mode decomposition according to an embodiment of this application.

[0046] Figure 4 The diagram shown is a flowchart illustrating a single decomposition process in the execution of a two-dimensional local mean decomposition according to an embodiment of this application.

[0047] Figure 5 The diagram shows a single extraction process in the extraction of product function components according to an embodiment of this application.

[0048] Figure 6 The diagram shown is a flowchart illustrating a current enhanced feature map acquisition method as described in an embodiment of this application.

[0049] Figure 7 The diagram shown is a structural schematic of a terminal as described in an embodiment of this application.

[0050] Explanation of reference numerals in the attached figures

[0051] 60: Terminal; 61: Processor; 62: Memory; 621: Operating System; 622: Application Program; 63: User Interface; 64: Network Interface; 65: Bus System. Detailed Implementation

[0052] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0053] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0054] To improve the accuracy of monitoring results for contaminated sites, it is usually necessary to remove spurious changes in images caused by natural environmental factors such as weather and lighting, so as to avoid these spurious changes affecting the detection of changes in the images and leading to inaccurate monitoring results.

[0055] Based on this, the following embodiments of this application provide a method, terminal and medium for monitoring soil contaminated sites based on remote sensing information. By separating color features, saturation channel data and brightness channel data are extracted and modal decomposition is performed to obtain each effective feature component, which is then fused with hue channel data to form an enhanced feature map. This not only effectively removes false changes in the image, but also enhances the features, thereby improving the accuracy of change detection and improving the monitoring effect of contaminated sites.

[0056] The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0057] like Figure 1 As shown, this embodiment provides a method for monitoring soil-contaminated sites, including:

[0058] S100: Acquire remote sensing data of the target plot at the current time and at the reference time, and perform preprocessing.

[0059] The target site refers to the corresponding area where pollution remediation or aggravation needs to be monitored.

[0060] The current time refers to the moment when the current situation of the target plot is monitored, while the reference time refers to the moment when the target plot is monitored. For example, the reference time is the initial moment when the target plot is monitored, or the moment when the target plot was last monitored.

[0061] Remote sensing data is acquired at the current time and at a reference time using equipment such as satellites or drones to reflect the status of the target plot at the current time and the reference time, thereby enabling the monitoring of the target plot.

[0062] Furthermore, the remote sensing data are preprocessed to align the spatial extent and resolution of the remote sensing data at the current time and the reference time, and to eliminate aberrations such as geometric distortion and spatial misalignment, thereby improving the accuracy of subsequent monitoring results.

[0063] For example, the preprocessing of remote sensing data includes, but is not limited to, geometric correction, radiometric coarse correction, image registration, etc. Those skilled in the art can adopt corresponding preprocessing methods according to actual needs, and this embodiment does not impose specific limitations.

[0064] S200 performs color feature separation on each remote sensing data to obtain the hue channel data, saturation channel data, and brightness channel data corresponding to the current time and the reference time, respectively.

[0065] Color feature separation refers to converting remote sensing data from RGB color space to HSV color space and extracting data from three channels—hue channel, saturation channel, and lightness channel—through channel separation.

[0066] Specifically, the hue channel data is extracted based on the remote sensing data at the current moment, serving as the current hue channel data; the saturation channel data is extracted based on the remote sensing data at the current moment, serving as the current saturation channel data; the lightness channel data is extracted based on the remote sensing data at the current moment, serving as the current lightness channel data; the hue channel data is extracted based on the remote sensing data at the reference moment, serving as the reference hue channel data; the saturation channel data is extracted based on the remote sensing data at the reference moment, serving as the reference saturation channel data; and the lightness channel data is extracted based on the remote sensing data at the reference moment, serving as the reference lightness channel data. Those skilled in the art should understand the principles and specific implementation methods of converting RGB color space to HSV color space and separating the data of each channel; these will not be elaborated upon in this embodiment.

[0067] It should be noted that the hue channel data in remote sensing data is used to characterize the color category information of ground objects, such as dominant wavelength and color attributes. It is less affected by natural environmental factors such as weather and lighting, and can significantly reflect the pollution status of the target area. Based on this, in the subsequent processing of this embodiment, the hue channel data is separated, and only the saturation and brightness channels in the remote sensing data are subjected to further modal decomposition operations to remove the pseudo-variations caused by natural environmental factors such as weather and lighting in the remote sensing data, retain and enhance the features that reflect the pollution status of the target area, thereby helping to improve the accuracy of monitoring the pollution status of the target area.

[0068] Furthermore, both the current hue channel data and the reference hue channel data undergo smoothing and denoising processing to suppress noise and improve data quality. This facilitates subsequent acquisition of the pollution status of the target plot based on the current hue channel data and the reference hue channel data, further enhancing the accuracy of monitoring the pollution status of the target plot. For example, the methods for smoothing and denoising the current hue channel data and the reference hue channel data include, but are not limited to, Gaussian filtering, median filtering, etc. Those skilled in the art can implement corresponding methods for smoothing and denoising according to actual needs; this embodiment does not impose specific limitations.

[0069] S300 performs mode decomposition operations based on each saturation channel data and each brightness channel data to obtain several currently effective feature components and several baseline effective feature components.

[0070] Among them, each currently valid feature component is used to characterize various pollution information of the target plot at the current time; each baseline valid feature component is used to characterize various pollution information of the target plot at the baseline time. It should be noted that the pollution information of the target plot is reflected through features such as image edges, texture, and surface disturbance of the image. That is, each currently valid feature component includes image edge, texture, and surface disturbance feature components extracted based on remote sensing data at the current time, and each baseline valid feature component includes image edge, texture, and surface disturbance feature components extracted based on remote sensing data at the baseline time.

[0071] Furthermore, modal decomposition operations are performed on the current saturation channel data or the current brightness channel data respectively to obtain each current effective feature component; modal decomposition operations are also performed on the reference saturation channel data or the reference brightness channel data respectively to obtain each reference effective feature component.

[0072] In some optional implementations, the modal decomposition operation is performed in the same way for any of the current saturation channel data, current brightness channel data, reference saturation channel data, and reference brightness channel data. Therefore, for the convenience of those skilled in the art, the current saturation channel data, current brightness channel data, reference saturation channel data, or reference brightness channel data for which modal decomposition operation needs to be performed will all be referred to as data to be processed, in order to specifically illustrate the specific execution method of the modal decomposition operation.

[0073] For example, such as Figure 2 As shown, the mode decomposition operation includes:

[0074] S310 performs two-dimensional empirical mode decomposition on the data to be processed, obtaining several mid-frequency intrinsic mode function components, several low-frequency intrinsic mode function components, and global residual components.

[0075] Two-dimensional empirical mode decomposition is used to perform multi-scale stratification of the data to be processed, so as to separate noise information, subject information and background information in the data to be processed. This helps to remove the pseudo-change parts in the data to be processed and facilitates the subsequent addition of features reflecting the pollution status of the target plot, thereby improving the accuracy of change detection and achieving better monitoring results.

[0076] In some alternative implementations, the two-dimensional empirical mode decomposition is a cyclically executed decomposition process, exemplarily, such as... Figure 3 As shown, a single decomposition process in the execution of two-dimensional empirical mode decomposition includes:

[0077] S311, obtain the current data to be decomposed and the current number of decompositions, and in combination with the preset local region size, obtain each local maximum point and each local minimum point.

[0078] Here, the data to be decomposed is the data used for decomposition in the current decomposition process, and the current decomposition count represents the number of times the decomposition process is executed in a loop. It should be noted that, at the initial execution of the decomposition, the data to be decomposed is the data to be processed.

[0079] Local maxima refer to the maxima within a local region, i.e., the points with the largest eigenvalues ​​within that local region. Local minima refer to the minima within a local region, i.e., the points with the smallest eigenvalues ​​within that local region. Eigenvalues ​​are numerical values ​​used to characterize the magnitude of attributes in the data to be decomposed. Specifically, when the data to be processed is the current saturation channel data or the baseline saturation channel data, the eigenvalue is the saturation value of each pixel; when the data to be processed is the current brightness channel data or the baseline brightness channel data, the eigenvalue is the brightness value of each pixel.

[0080] In the current data to be decomposed, each region of a preset local region size is extracted, and these are called local regions. To facilitate understanding of the process of extracting local maxima and local minima, a specific example will be used below.

[0081] Taking a preset local region size of 3*3 pixels as an example, any 3*3 pixel region is extracted from the current data to be decomposed as a local region, and the maximum and minimum values ​​within this local region are obtained. Specifically, when the data to be processed is the current saturation channel data or the reference saturation channel data, the pixel with the largest saturation value in the local region is obtained as the maximum value, and the pixel with the smallest saturation value in the local region is obtained as the minimum value; when the data to be processed is the current brightness channel data or the reference brightness channel data, the pixel with the largest brightness value in the local region is obtained as the maximum value, and the pixel with the smallest brightness value in the local region is obtained as the minimum value. This process is repeated for the entire current data to be decomposed, obtaining the maximum and minimum values ​​in all local regions to serve as the local maximum and local minimum values.

[0082] S312, based on each local maximum point, fit to obtain the upper envelope surface; and based on each local minimum point, fit to obtain the lower envelope surface.

[0083] Specifically, based on the eigenvalue magnitudes of local maxima, a surface representing the changes in eigenvalue magnitudes is fitted as the upper envelope surface; and based on the eigenvalue magnitudes of local minima, a surface representing the changes in eigenvalue magnitudes is fitted as the lower envelope surface. It should be noted that the fitting method for the upper envelope surface based on each local maxima and the fitting method for the lower envelope surface based on each local minima can employ two-dimensional cubic spline interpolation. Those skilled in the art can select appropriate fitting methods to obtain the upper and lower envelope surfaces according to actual needs; this embodiment does not impose specific limitations.

[0084] S313, obtain the local mean surface based on the upper and lower envelope surfaces; obtain the difference between the current data to be decomposed and the local mean surface as the current first residual data, and calculate the standard deviation of the current first residual data and the current data to be decomposed to determine whether the standard deviation is within the stopping interval.

[0085] Specifically, based on each point on the upper and lower envelope surfaces, the average value is calculated to serve as the points of the local mean surface, thereby obtaining the local mean surface. The calculation method is as follows:

[0086]

[0087] in, It is a local mean surface. For the upper envelope, This is the lower envelope.

[0088] Furthermore, the difference between the current data to be decomposed and the local mean surface is calculated and used as the first residual data. The calculation method is as follows:

[0089]

[0090] in, This is the first residual data. This is the data to be decomposed.

[0091] Furthermore, calculate the standard deviation of the current first residual data and the current data to be decomposed. The calculation method is as follows:

[0092]

[0093] in, The standard deviation is denoted as .

[0094] The stopping interval is used to reflect the stopping condition of the cyclic decomposition process in two-dimensional empirical mode decomposition. That is, when the standard deviation is within the stopping interval, the cyclic decomposition process of two-dimensional empirical mode decomposition stops; otherwise, the next decomposition process continues. It should be noted that those skilled in the art can set the specific range of the stopping interval according to actual needs, and this embodiment does not make a specific limitation. For example, the stopping interval is [0.2, 0.3].

[0095] S314, if yes, output the current first residual data as the intrinsic mode function, output the current decomposition count as the corresponding decomposition count, output the difference between the data to be processed and the current first residual data as the global residual component, and stop the decomposition.

[0096] S314', if not, output the current first residual data as the intrinsic mode function component, output the current decomposition count as the corresponding decomposition count, increment the decomposition count by one as the new current decomposition count, use the intrinsic mode function component as the new current data to be decomposed, and re-execute the next decomposition process.

[0097] It should be noted that the intrinsic mode functions (IMFs) include high-frequency IMFs, mid-frequency IMF components, and low-frequency IMF components. The high-frequency IMFs reflect the noise information in the data to be processed, the mid-frequency IMF components reflect the main information in the data, and the low-frequency IMF components, together with the global residual components, reflect the background information in the data.

[0098] Based on this, in this embodiment, in order to remove the spurious changes in the data to be processed, the high-frequency intrinsic mode functions in each intrinsic mode function are eliminated, while the mid-frequency intrinsic mode function components, low-frequency intrinsic mode function components, and global residual components are retained.

[0099] Specifically, for each intrinsic mode function (IMF) component, its corresponding decomposition number is obtained. If the decomposition number is no more than 3, the IMF component is a high-frequency IMF and is discarded. If the decomposition number exceeds 3, the dominant wavelength corresponding to the IMF component is obtained through Fourier transform. If the dominant wavelength is in the mid-frequency wavelength range, the IMF component is a mid-frequency IMF component; if the dominant wavelength is in the low-frequency wavelength range, the IMF component is a low-frequency IMF component. It should be noted that those skilled in the art should know the specific method of obtaining the dominant wavelength through Fourier transform, which will not be elaborated here.

[0100] Furthermore, those skilled in the art can set the corresponding mid-frequency wavelength range and low-frequency wavelength range according to actual needs, and this embodiment does not impose specific limitations. For example, the mid-frequency wavelength range is [20m, 500m], and the low-frequency wavelength range is (500, +∞).

[0101] S320 performs two-dimensional local mean decomposition on each mid-frequency intrinsic mode function component to obtain several product function components and local residuals corresponding to each mid-frequency intrinsic mode function component.

[0102] Two-dimensional local mean decomposition (LMD) is used to enhance the features of each mid-frequency intrinsic mode function (IMF) component, thereby strengthening local fine-grained features. It should be noted that the IMF components reflect the main information in the data to be processed, including information on soil tillage, building demolition and construction, and vegetation changes within the site. Based on this, LMD enhances the main information of each IMF component, facilitating subsequent detection of changes in the site from the baseline time to the current time, thus improving the accuracy of contaminated site monitoring.

[0103] In some alternative implementations, the two-dimensional local mean decomposition is also a cyclically executed decomposition process, for example, such as... Figure 4 As shown, for any mid-frequency intrinsic mode function component, the single decomposition process in the two-dimensional local mean decomposition includes:

[0104] S321: Based on the current intrinsic mode function components to be decomposed, extract the corresponding product function components and output them.

[0105] Here, the eigenmode function component to be decomposed is the component used for decomposition in the current decomposition process. It should be noted that, during the initial decomposition, the eigenmode function component to be decomposed is the mid-frequency eigenmode function component corresponding to the two-dimensional local mean decomposition process.

[0106] The product function component is used to characterize information such as texture oscillation direction, periodic structure, edge orientation, and local spatial frequency in the intrinsic mode function component, thereby reflecting the local fine-grained features in the intrinsic mode function component, so as to facilitate feature enhancement.

[0107] Optionally, the extraction of the product function components is actually a loop-based extraction process. Specifically, such as... Figure 5 As shown, a single extraction process includes:

[0108] S3211, obtain the current instantaneous amplitude and the current demodulated signal; and based on the current demodulated signal, obtain the corresponding component local mean surface and component local envelope surface.

[0109] The current instantaneous amplitude is used to reflect the energy of the product function components, and the current demodulated signal is used to characterize the oscillation of the product function components.

[0110] Specifically, for the current demodulated signal, based on the preset size of the local region of each component, the local maxima and local minima of each component are obtained; and based on the local maxima of each component, the upper envelope of the component is fitted and obtained, and based on the local minima of each component, the lower envelope of the component is fitted and obtained, thereby obtaining the local mean surface of the component. It should be noted that the methods for obtaining the upper and lower envelopes of the components are described in steps S311 to S312 above, and will not be repeated here in this embodiment.

[0111] Furthermore, based on each point on the upper and lower envelopes of the component, the average value is calculated to serve as the points of the component's local mean surface, thereby obtaining the component's local mean surface. The calculation method is as follows:

[0112]

[0113] in, For the local mean surface of the component, For the component envelope, The envelope of the component.

[0114] Furthermore, based on each point on the upper and lower envelopes of the component, half of the difference is calculated to serve as the points on the local envelope of the component. The calculation method is as follows:

[0115]

[0116] in, This is the local envelope of the component.

[0117] S3212, obtain the difference between the current demodulated signal and the component local mean surface, and normalize it based on the component local envelope surface to obtain a new demodulated signal.

[0118] Specifically, the difference between the current intrinsic mode function component to be decomposed and the local mean surface of the component is calculated, and then divided by the local envelope surface of the first component. The quotient is used as the new demodulated signal. The calculation method is as follows:

[0119]

[0120]

[0121] in, For the new demodulated signal, These are the components of the intrinsic mode function to be decomposed.

[0122] S3213, determine whether the local envelope of a component is constant at one.

[0123] S3214, if so, output the product of the component's local envelope, the current instantaneous amplitude, and the new demodulated signal as the product function component, and stop the extraction process.

[0124] S3214', if not, take the product of the component local envelope and the current instantaneous amplitude as the new current instantaneous amplitude, and take the new demodulated signal as the current demodulated signal, and re-execute the next extraction process.

[0125] S322: Obtain the current local residual component based on the difference between the current eigenmode function component to be decomposed and the product function component, and determine whether the current local residual component satisfies the monotonicity condition.

[0126] The monotonicity condition is used to characterize the conditions that the current local residual component must satisfy when it becomes the final local residual component. For example, the monotonicity condition is that the current local residual component is monotonically increasing or monotonically decreasing in all directions, that is, the current local residual component has only one maximum and one minimum value.

[0127] Specifically, those skilled in the art should know how to determine whether the current local residual components are monotonically increasing or monotonically decreasing in all directions, which will not be elaborated here in this embodiment.

[0128] S323, if satisfied, then the current local residual component will be output as the local residual component.

[0129] It should be noted that when the current local residual components are monotonically increasing or monotonically decreasing in all directions, that is, the information such as the texture oscillation direction, periodic structure, edge direction, and local spatial frequency of the mid-frequency intrinsic mode function components currently being decomposed has been extracted into each product function component. Therefore, the remaining current local residual components are no longer decomposed and are output as local residual components.

[0130] If S323' is not satisfied, the current local residual component is taken as the new current eigenmode function component to be decomposed, and the next decomposition process is executed again.

[0131] Based on this, several product function components and local residuals corresponding to each mid-frequency intrinsic mode function component can be effectively extracted.

[0132] S330 uses each product function component, each local residual, each low-frequency intrinsic mode function component, and the global residual component as each effective characteristic component.

[0133] It should be noted that when the data to be processed is the current saturation channel data or the current brightness channel data, each product function component, each local residual, each low-frequency intrinsic mode function component, and the global residual component are all used as the current effective feature components; when the data to be processed is the reference saturation channel data or the reference brightness channel data, each product function component, each local residual, each low-frequency intrinsic mode function component, and the global residual component are all used as the reference effective feature components.

[0134] S400 performs feature fusion based on each current effective feature component and each benchmark effective feature component, combined with the corresponding tone channel data, to obtain the current enhanced feature map and the benchmark enhanced feature map.

[0135] The current enhanced feature map is a feature map fused with the current effective feature components and the current hue channel data features, to enhance the reflection of the pollution status of the target plot at the current time. The baseline enhanced feature map is a feature map fused with the baseline effective feature components and the baseline hue channel data features, to enhance the reflection of the pollution status of the target plot at the baseline time.

[0136] The current enhanced feature map and the baseline enhanced feature map are obtained in the same way, specifically, as follows: Figure 6 As shown, taking the acquisition of the current enhanced feature map based on each currently effective feature component and combined with the current tone channel data as an example, the acquisition methods include:

[0137] S410, based on the current effective feature components corresponding to the current saturation channel data, the current effective feature components corresponding to the current brightness channel data, and the current hue channel data, respectively, fuses to obtain the current multi-channel feature map.

[0138] Specifically, based on the current effective feature components corresponding to the current saturation channel data, the current effective feature components corresponding to the current brightness channel data, and the current hue channel data, a three-channel feature map is stitched together, with the three channels representing brightness, saturation, and hue, respectively. Specifically, the current effective feature components corresponding to the current saturation channel data are fused to form effective saturation channel data, the current effective feature components corresponding to the current brightness channel data are fused to form effective brightness channel data, and the effective saturation channel data, effective brightness channel data, and current hue channel data are stitched together to form a three-channel feature map. It should be noted that those skilled in the art should understand the specific execution methods and principles of feature fusion and channel stitching, which will not be elaborated upon in this embodiment.

[0139] S420, calculate the energy value and gradient value of each current effective feature component to obtain the weight corresponding to each current effective feature component.

[0140] Among them, the energy value is used to characterize the information richness of the corresponding current effective feature component. For example, the energy value is the local variance or the sum of squared gradient magnitudes; the gradient value is used to characterize the structural clarity of the corresponding current effective feature component. For example, the gradient value is the gradient magnitude.

[0141] Based on this, for any currently valid feature component, the sum of the energy value and gradient value is calculated and normalized to serve as the weight of that currently valid feature component.

[0142] It should be noted that those skilled in the art can select the corresponding parameter values ​​as the energy value and gradient value of the current effective feature component according to actual needs, and perform calculations. This embodiment does not impose specific limitations here.

[0143] S430 performs convolutional weighting on the current multi-channel feature map based on the weights corresponding to each current effective feature component to obtain the current enhanced feature map.

[0144] For each currently effective feature component in the multi-channel feature map, the corresponding part is weighted based on its weight and then convolved to form an image, which is used as the current enhanced feature map.

[0145] Based on this, the current enhanced feature image can be obtained based on each currently effective feature component and the current brightness channel data.

[0146] S500 performs change detection and annotation based on the current enhanced feature map and the baseline enhanced feature map to obtain the pollution changes of the target plot.

[0147] For example, a pre-trained Transformer change detection network is used for change detection. The current enhanced feature map and the baseline enhanced feature map are input for detection to identify and label changes in the target plot at the current time relative to the baseline time. The Transformer change detection network can employ an architecture coupling a dual-branch symmetric encoder and a Transformer decoder. This architecture performs block embedding and positional encoding operations on the input current and baseline enhanced feature maps, respectively, to transform them into sequential feature vectors with spatial location information. These two sets of feature vectors are then fed in parallel into a Transformer encoder composed of alternating stacked multi-head self-attention layers and feedforward neural network layers. The self-attention mechanism establishes long-distance dependencies within and between the two temporal features to detect changes in the image.

[0148] Optionally, enhanced feature maps of the same plot of land at different times are obtained, and a labeling tool is used to delineate the pollution variation range between images at different times pixel by pixel, generating accurate binary change mask labels. Then, the delineated images are divided into training and validation sets based on a preset ratio, and input into the Transformer change detection network for training. The accuracy of the Transformer change detection network is verified using the validation set to finally obtain the trained Transformer change detection network. Furthermore, before inputting the images and their labels into the Transformer change detection network, data augmentation processing such as random cropping, rotation, mirror flipping, and brightness / contrast perturbation is performed to improve sample diversity and suppress model overfitting.

[0149] Furthermore, based on the results detected by the Transformer change detection network, the change information of the current time relative to the reference time is obtained, and this change information is annotated back to the remote sensing data corresponding to the current time. This allows staff to intuitively obtain the change of the current time relative to the reference time based on the remote sensing data, thereby analyzing the change of pollution status of the current time relative to the reference time.

[0150] Based on the same technical concept, the soil pollution site monitoring method based on remote sensing information provided in the embodiments of the present invention can be implemented on the terminal side or the server side.

[0151] like Figure 7 The diagram illustrates an optional hardware structure of a terminal according to an embodiment of the present invention. The terminal 60 can be a mobile phone, computer device, tablet device, personal digital processing device, factory back-end processing device, etc. The terminal 60 includes at least one processor 61, a memory 62, at least one network interface 64, and a user interface 63. The various components in the device are coupled together via a bus system 65. It is understood that the bus system 65 is used to realize communication between these components. In addition to a data bus, the bus system 65 also includes a power bus, a control bus, and a status signal bus.

[0152] The user interface 63 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0153] It is understood that memory 62 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory characterized in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable categories of memory.

[0154] In this embodiment of the invention, memory 62 is used to store various types of data to support terminal operation. Examples of this data include: any executable program for operation on terminal 60, such as operating system 621 and application program 622; operating system 621 includes various system programs, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. Application program 622 may include various applications, such as media player, browser, etc., for implementing various application services. The method for monitoring soil contaminated sites based on remote sensing information provided in this embodiment of the invention can be included in application program 622.

[0155] The methods disclosed in the above embodiments of the present invention can be applied to processor 61, or implemented by processor 61. Processor 61 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 61 or by instructions in the form of software. The processor mentioned above may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 61 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. Processor 61 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0156] In an exemplary embodiment, terminal 60 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to execute the aforementioned method.

[0157] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when invoked by a processor, implements the soil contamination monitoring method based on remote sensing information provided by this invention.

[0158] Computer-readable storage media can be tangible devices capable of holding and storing instructions used by an instruction execution device. Computer-readable storage media can be, for example, (but not limited to) electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, and mechanical encoding devices.

[0159] The computer-readable program represented herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network, to an external computer or external storage device. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards these instructions to the computer-readable storage medium in the respective computing / processing device.

[0160] In summary, this application separates color features and extracts saturation and brightness channel data for modal decomposition to obtain effective feature components. These components are then fused with hue channel data to form an enhanced feature map. This not only effectively removes false changes in the image but also enhances features, thereby improving the accuracy of change detection and enhancing the monitoring effect of contaminated sites.

[0161] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0162] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for monitoring soil-contaminated sites based on remote sensing information, comprising: The remote sensing data of the target plot at the current time and the reference time are acquired and preprocessed respectively; Color feature separation is performed based on the preprocessed remote sensing data to obtain the current hue channel data, current saturation channel data, and current brightness channel data corresponding to the current time, as well as the reference hue channel data, reference saturation channel data, and reference brightness channel data corresponding to the reference time. Modal decomposition operations are performed based on the current saturation channel data and the current brightness channel data to obtain several current effective feature components; and modal decomposition operations are performed based on the reference saturation channel data and the reference brightness channel data to obtain several reference effective feature components. Based on each of the currently effective feature components, and combined with the current tone channel data, feature fusion is performed to obtain the current enhanced feature map; Based on each of the aforementioned effective feature components, and combined with the aforementioned reference tone channel data, feature fusion is performed to obtain a reference enhanced feature map; Change detection is performed based on the current enhanced feature map and the baseline enhanced feature map, and the changes are marked to obtain the pollution changes of the target plot; For any one of the current saturation channel data, the current brightness channel data, the reference saturation channel data, and the reference brightness channel data, the modal decomposition operation includes: Two-dimensional empirical mode decomposition is performed on the data to be processed to obtain several mid-frequency intrinsic mode function components, several low-frequency intrinsic mode function components, and global residual components; wherein, the data to be processed is any one of the current saturation channel data, the current brightness channel data, the reference saturation channel data, and the reference brightness channel data; Based on the two-dimensional local mean decomposition of each of the mid-frequency intrinsic mode function components, several product function components and local residuals corresponding to each of the mid-frequency intrinsic mode function components are obtained; Each of the product function components, each of the local residuals, each of the low-frequency intrinsic mode function components, and the global residual components are used as each of the effective feature components; wherein, when the data to be processed is the current saturation channel data or the current brightness channel data, each of the product function components, each of the local residuals, each of the low-frequency intrinsic mode function components, and the global residual components are all used as each of the current effective feature components; when the data to be processed is the reference saturation channel data or the reference brightness channel data, each of the product function components, each of the local residuals, each of the low-frequency intrinsic mode function components, and the global residual components are all used as each of the reference effective feature components; For any of the aforementioned mid-frequency intrinsic mode function components, a two-dimensional local mean decomposition is performed, which is a cyclic decomposition process. Each decomposition cycle includes: Based on the current intrinsic mode function components to be decomposed, extract the corresponding product function components and output them; The current local residual component is obtained based on the difference between the current eigenmode function component to be decomposed and the product function component. If the current local residual component satisfies the monotonicity condition, the current local residual component is output as the local residual component. Otherwise, the current local residual component is used as the new current eigenmode function component to be decomposed, and the next decomposition process is executed again.

2. The method according to claim 1, characterized in that, The process of performing two-dimensional empirical mode decomposition on the data to be processed to obtain several mid-frequency intrinsic mode function components, several low-frequency intrinsic mode function components, and global residual components is a cyclic decomposition process. A single decomposition process includes: Obtain the current data to be decomposed and the current number of decompositions, and combine them with the preset local region size to obtain each local maximum point and each local minimum point; Based on each of the local maxima, the upper envelope is fitted and obtained; and based on each of the local minima, the lower envelope is fitted and obtained. Based on the upper and lower envelope surfaces, a local mean surface is obtained. The difference between the current data to be decomposed and the local mean surface is obtained as the current first residual data. The standard deviation of the current first residual data and the current data to be decomposed is calculated. If the standard deviation is within the stopping interval, the current first residual data is output as an intrinsic mode function, the current decomposition count is output as the corresponding decomposition count, the difference between the data to be processed and the current first residual data is output as the global residual component, and the decomposition stops. Otherwise, the current first residual data is output as the intrinsic mode function component, the current decomposition count is output as the corresponding decomposition count, the decomposition count is incremented by one as the new current decomposition count, the intrinsic mode function component is used as the new current data to be decomposed, and the next decomposition process is executed again. Specifically, for each intrinsic mode function component, the corresponding dominant wavelength is obtained through Fourier transform. If the number of decompositions corresponding to any intrinsic mode function component exceeds a preset threshold and the corresponding dominant wavelength is within the mid-frequency wavelength range, then the intrinsic mode function component is a mid-frequency intrinsic mode function component. If the number of decompositions corresponding to any intrinsic mode function component exceeds a preset threshold and the corresponding dominant wavelength is within the low-frequency wavelength range, then the intrinsic mode function component is a low-frequency intrinsic mode function component.

3. The method according to claim 1, characterized in that, The extraction of the product function components is a cyclic extraction process, and a single extraction process includes: The system acquires the current instantaneous amplitude and the current demodulated signal; based on the current demodulated signal and a preset local region size for each component, it acquires the local maxima and local minima for each component; based on the local maxima, it fits and acquires the upper envelope of the component; based on the local minima, it fits and acquires the lower envelope of the component; based on the upper and lower envelopes, it acquires the corresponding local mean and local envelope of the component. The difference between the current demodulated signal and the component local mean surface is obtained, and normalized based on the component local envelope surface to obtain a new demodulated signal. If the component local envelope surface is constant at one, the product of the component local envelope surface, the current instantaneous amplitude, and the new demodulated signal is used as the product function component output, and the extraction process stops. Otherwise, the product of the component local envelope surface and the current instantaneous amplitude is used as the new current instantaneous amplitude, and the new demodulated signal is used as the new current demodulated signal, and the next extraction process is executed again.

4. The method according to claim 1, characterized in that, The step of performing feature fusion based on each of the currently effective feature components and the current tone channel data to obtain the current enhanced feature map includes: The current multi-channel feature map is obtained by fusing the current effective feature components corresponding to the current saturation channel data, the current effective feature components corresponding to the current brightness channel data, and the current hue channel data respectively. Calculate the energy value and gradient value of each of the current effective feature components to obtain the weight corresponding to each of the current effective feature components; Based on the weights corresponding to each of the current effective feature components, the multi-channel feature map is convolved and weighted to obtain the current enhanced feature map. The step of performing feature fusion based on each of the aforementioned effective feature components and combining the aforementioned hue channel data to obtain a reference enhanced feature map includes: The reference multi-channel feature map is obtained by fusing the reference effective feature components corresponding to the reference saturation channel data, the reference effective feature components corresponding to the reference brightness channel data, and the reference hue channel data respectively. Calculate the energy value and gradient value of each of the aforementioned effective feature components to obtain the weights corresponding to each of the aforementioned effective feature components; Based on the weights corresponding to each of the aforementioned effective feature components, the multi-channel feature map is convolved and weighted to obtain the benchmark enhanced feature map.

5. The method according to claim 1, characterized in that, The process of detecting and labeling changes based on the current enhanced feature map and the baseline enhanced feature map to obtain the pollution changes of the target plot includes: Based on a pre-trained change detection network, the current enhanced feature map and the baseline enhanced feature map are input to divide each region and obtain the corresponding annotation information for each region; The current enhanced feature map is restored to obtain an RGB image, and the RGB image is annotated in combination with the annotation information corresponding to each region, so as to output an annotated RGB image reflecting the pollution change of the target plot.

6. The method according to claim 1, characterized in that, After obtaining the current tone channel data and the reference tone channel data, the method further includes performing smoothing and noise reduction processing on both the current tone channel data and the reference tone channel data.

7. A terminal, characterized in that, include: A processor and a memory, wherein the memory and the processor are communicatively connected; The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to enable the terminal to perform the soil contaminated site monitoring method as described in any one of claims 1 to 6.

8. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the soil contaminated site monitoring method as described in any one of claims 1 to 6.

Images