Deep learning based remote sensing data intelligent denoising and enhancement system and method

Abstract

The application discloses a remote sensing data intelligent denoising and enhancing system and method based on deep learning, and belongs to the technical field of image processing and data analysis.The application comprises the following steps: S1, data acquisition, remote sensing images of a target marine area are collected through satellite and unmanned aerial vehicle sensors; S2, an intelligent denoising model is built, feature training is conducted through a deep learning network, and feature denoising processing is conducted on the obtained remote sensing image data; S3, image registration and merging are conducted on the images whose features have been denoised; S4, image post-processing is conducted on the images whose image registration and merging have been completed; and S5, result visual display and data analysis feedback are conducted, and the processed images are visually displayed through image processing software.The application directly learns and summarizes noise features through the use of a deep learning method, does not cause training complication due to excessive noise, greatly saves resources, and shortens the training time.

Description

Technical Field

[0001] This invention relates to the field of image processing and data analysis technology, specifically to a system and method for intelligent noise reduction and enhancement of remote sensing data based on deep learning. Background Technology

[0002] Image intelligent noise reduction technology is a technique that uses artificial intelligence and deep learning algorithms to reduce noise and artifacts in images and improve image quality.

[0003] In remote sensing image processing, especially in the treatment of noise in remote sensing images, noise has a significant impact on the image presentation and quality. Currently, noise processing for remote sensing images mainly uses Gaussian filtering, mean filtering, and bilateral filtering. If the image blurring in a single image is caused by Gaussian noise, then using Gaussian filtering for denoising will yield good results. However, in many cases, there are multiple images and multiple types of noise, which may be caused by various factors. Current machine learning-based denoising models rely on image datasets for self-learning, summarizing various noise features through a large amount of image data. In this case, it requires a long training time with a large number of images, resulting in wasted resources and preventing widespread application. Moreover, the denoising model becomes more complex during training due to excessive noise, leading to poor denoising results. Summary of the Invention

[0004] The purpose of this invention is to provide a system and method for intelligent noise reduction and enhancement of remote sensing data based on deep learning, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a deep learning-based intelligent noise reduction and enhancement method for remote sensing data, comprising the following steps: S1. Data acquisition: Collect remote sensing images of the target ocean area in different bands through satellite and UAV sensors; including visible light images, near-infrared images and short-wave infrared images; S2. Build an intelligent noise reduction model, train features through a deep learning network, and perform feature noise reduction processing on the acquired remote sensing image data to achieve intelligent noise reduction. S3. Perform image registration and merging on images of different bands that have undergone feature denoising to enhance the detail display of remote sensing images; S4. Perform image post-processing on the images that have been registered and merged, including color correction and image enhancement, to reduce color difference in remote sensing images and achieve image enhancement. S5. Results visualization and data analysis feedback: The processed images are visualized using image display software, making it easy for users to analyze and interpret them; the image data is analyzed in conjunction with specific application scenarios, and the analysis results are fed back to the intelligent noise reduction model to optimize the model.

[0006] The image registration and merging process includes image registration, radiometric correction, and normalization. Image registration spatially registers images of different bands, aligning them in the same coordinate system to ensure a one-to-one correspondence between pixels of images of different bands. Radiometric correction performs radiometric correction on the images to eliminate radiometric differences introduced by sensor and atmospheric factors. Normalization standardizes pixel values ​​to the same range.

[0007] In step S1, Satellite sensors offer wide coverage and provide multispectral, high-resolution data, making them suitable for long-term monitoring of large areas. Unmanned aerial vehicle (UAV) sensors offer adjustable flight altitude and high flexibility, making them suitable for detailed monitoring of localized areas, and for high-precision monitoring and rapid response in specific regions. The visible light images reflect the color and reflectivity of the ocean surface and are used to detect ocean surface pollution and algae cover; the near-infrared images reflect the infrared reflectivity of vegetation and water bodies and are used to monitor the distribution and changes of marine vegetation and plankton; the short-wave infrared images have strong penetration and can provide information about water bodies and wetlands, and are used to monitor changes in ocean temperature, humidity and chemical composition.

[0008] In step S2, the feature training includes statistical analysis features, frequency domain analysis features, autocorrelation features, and histogram features; Statistical analysis characteristics include: Mean:

[0009] Where Mean represents the mean value of pixels in the image; K represents the total number of pixels in the image, z i This represents the pixel value of point i in the image; Variance:

[0010] Wherein, Variance represents the variance of the image pixel values; Standard Deviation:

[0011] Wherein, Standard Deviation represents the standard deviation of image pixel values; Kur:

[0012] Where Kur represents the kurtosis of the image; The kurtosis reflects the sharpness of the image brightness distribution; Frequency domain analysis features: Extracting frequency information from an image by performing a frequency domain transformation on the image; Fourier transform: transforms an image from the spatial domain to the frequency domain to analyze the frequency components of the image;

[0013] Where F(u,v) represents the image in the frequency domain; f(x,y) represents the image in the spatial domain; M represents the size of the image in the x-direction; N represents the size of the image in the y-direction; u represents the horizontal frequency component in the frequency domain; v represents the vertical frequency component in the frequency domain; and x and y represent the pixel coordinates in the spatial domain. Power spectral density E: describes the energy distribution of different frequency components in an image;

[0014] Where E(u,v) represents the power of each frequency component in the frequency domain; |F(u,v)| represents the amplitude of each frequency component in the frequency domain. High-frequency and low-frequency components of the image are extracted using a bandpass filter. High-frequency components: satisfying u≥u c , v≥v c u c It is half of the horizontal frequency range; v c It is half of the vertical frequency range; High-frequency component energy E high :

[0015] Total energy E total :

[0016] High-frequency component energy ratio (HF):

[0017] The high-frequency and low-frequency components of the image reflect the edge and texture features of the image as well as the overall smoothness of the image. Autocorrelation characteristics: The autocorrelation function is used to describe the correlation between pixels in an image;

[0018] Where K represents the total number of pixels in the image; R(Δx,Δy) represents the autocorrelation value between pixels at the displacement (Δx,Δy); and f(x+Δx,y+Δy) represents the image after displacement. Histogram characteristics: Histogram mean H Mean :

[0019] Among them, H Mean Let L represent the mean of the histogram; L represent the number of gray levels; p(t) represent the probability of gray level t. Histogram variance H var :

[0020] Among them, H var Indicates the variance of the histogram; Feature training is performed on the intelligent noise reduction model.

[0021] In step S2, noise category C k Including Gaussian noise, impulse noise, Poisson noise, and shot noise, the specific method for identification is as follows: In statistical features, kurtosis (Kur) is selected as the parameter:

[0022] Kur represents kurtosis; Indicates noise type C k The average value of the lower kurtosis; Indicates noise type C k The standard deviation below; Selecting the energy ratio of high-frequency components in the frequency domain characteristics:

[0023] Wherein, HF represents the energy ratio of the high-frequency components; Indicates noise type C k The average value of the energy ratio of the lower high-frequency components; Indicates noise type C k Standard deviation of the energy ratio of lower high-frequency components; Selecting the zero-delay autocorrelation value as a parameter from the autocorrelation features: the zero-delay autocorrelation value can be obtained by setting the values ​​of Δx and Δy to 0.

[0024] Where AC represents the zero-delay autocorrelation value; Indicates noise type C k The average value of zero-delay autocorrelation; Indicates noise type C kThe standard deviation of the zero-delay autocorrelation value; Select the histogram mean as a parameter in the histogram features:

[0025] Among them, H var Indicates the variance of the histogram; Indicates noise type C k The average variance of the histogram below; Indicates noise type C k The standard deviation of the variance of the histogram below; Calculate the probability:

[0026]

[0027]

[0028]

[0029]

[0030] Among them, P(Kur), P(HF), P(AC) and P(H var P(C) represents the marginal probability of the feature; k () indicates noise type C k The prior probability, i.e., the probability that no noise feature was observed; P(Kur|C k ) indicates that in noise type C k Below, the probability density function of kurtosis; P(HF|C k ) indicates that in noise type C k Below, the probability density function of the energy ratio of high-frequency components; P(AC|C k ) indicates that in noise type C k Below, the probability density function of the zero-delay autocorrelation value; P(H var |C k ) indicates that in noise type C k Below, the probability density function of the histogram mean;

[0031] in, The noise type is then verified, including Gaussian noise, impulse noise, Poisson noise, and shot noise. The posterior probability of each noise type is calculated, and the noise type with the highest posterior probability is selected.

[0032] The Gaussian noise is reduced using wavelet transform, and the specific steps are as follows: S5-1. Decompose the image into different scales and directions through wavelet transform, and perform thresholding on the wavelet coefficients. S5-2. Perform wavelet transform on the image to obtain wavelet coefficients; perform soft thresholding and hard thresholding on the wavelet coefficients. S5-3. Perform inverse wavelet transform on the processed wavelet coefficients to reconstruct the image.

[0033] The impulse noise is reduced using adaptive median filtering, and the specific steps are as follows: S6-1. Initialize window size; S6-2. Calculate the median, maximum, and minimum values ​​of the pixels within the window; S6-3. Adjust the window size based on whether the pixel value is between the median and extreme values; S6-4. Replace the center pixel with the median of the adaptive window.

[0034] The method for processing Poisson noise is to use Anscombe transform to convert an image containing Poisson noise into a Gaussian noise-like image, then use wavelet transform for noise reduction, and finally perform inverse transform.

[0035] The Anscombe transform is specifically as follows:

[0036] Where I' represents a Gaussian noise image; I represents a Poisson noise image; The inverse transformation is:

[0037] Shot noise is reduced using gamma filtering.

[0038] The image registration and merging adopts a weighted average merging method, specifically:

[0039] in, This represents the pixel value at (x, y) in the merged image; , and represents the pixel value at (x,y) of the visible light image, near-infrared image, and short-wave infrared image, respectively; w1, w2, and w3 represent the weights of the visible light image, near-infrared image, and short-wave infrared image, respectively; the weights are set according to the amount of information contained in the image based on the experimental conclusions, specifically w1=0.4, w2=0.3, and w3=0.3.

[0040] In step S4, the color correction and image enhancement specifically include: Color correction: Adjust the gain of each channel to make the image colors more balanced; select the white areas in the image and perform white balance correction to ensure that the white areas look more natural; use a reference image to perform color mapping so that the color distribution of the target image is consistent with the reference image; Image enhancement: Enhance image contrast to make bright and dark areas more distinct; perform unsharpening masking to enhance image edges and details, making the image clearer; convert to HSV color space and adjust color saturation and hue.

[0041] The system for intelligent denoising and enhancement of remote sensing data based on deep learning implements intelligent denoising and enhancement methods for remote sensing data based on deep learning. The system includes a data acquisition module, an intelligent denoising model module, an image registration and merging module, an image post-processing module, and a result display module. The data acquisition module is used to acquire remote sensing image data of the target area through satellite and UAV sensors; the intelligent noise reduction model module is used to process the acquired remote sensing images, remove various types of noise, including Gaussian noise, impulse noise, Poisson noise, and shot noise, and enhance image quality; the image registration and merging module is used to register and merge multispectral band images to generate a comprehensive image containing more information and details; the image post-processing module is used to further optimize image quality, enhance image details and contrast, and perform color correction and image enhancement; the result display module is used to display the processed image to the user, providing visualization tools and interfaces to support further analysis and operation of the image. The output of the data acquisition module is connected to the input of the intelligent noise reduction model module; the output of the intelligent noise reduction model module is connected to the input of the image registration and merging module; the output of the image registration and merging module is connected to the input of the image post-processing module; and the output of the image post-processing module is connected to the input of the result display module.

[0042] Compared with the prior art, the beneficial effects achieved by the present invention are: 1. This invention analyzes a large number of remote sensing images collected manually to obtain noise features of various images. Instead of the model summarizing features by itself through a large amount of image data, it directly learns the summarized features, which avoids excessive noise that complicates training, greatly saves resources, and shortens training time. 2. This invention applies different treatments to different types of noise, achieving better noise reduction results. Attached Figure Description

[0043] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram illustrating the steps of the intelligent noise reduction and enhancement method for remote sensing data based on deep learning, as described in this invention. Figure 2 It is a Gaussian noise histogram; Figure 3 It is an impulse noise histogram; Figure 4 It is a Poisson noise histogram; Figure 5 This is a shot noise histogram. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Please see Figures 1-5 This invention provides a technical solution: a method for intelligent noise reduction and enhancement of remote sensing data based on deep learning, which includes the following steps: S1. Data acquisition: Collect remote sensing images of the target ocean area in different bands through satellite and UAV sensors; including visible light images, near-infrared images and short-wave infrared images; S2. Build an intelligent noise reduction model, train features through a deep learning network, and perform feature noise reduction processing on the acquired remote sensing image data to achieve intelligent noise reduction. S3. Perform image registration and merging on images of different bands that have undergone feature denoising to enhance the detail display of remote sensing images; S4. Perform image post-processing on the images that have been registered and merged, including color correction and image enhancement, to reduce color difference in remote sensing images and achieve image enhancement. S5. Results visualization and data analysis feedback: The processed images are visualized using image display software, making it easy for users to analyze and interpret them; the image data is analyzed in conjunction with specific application scenarios, and the analysis results are fed back to the intelligent noise reduction model to optimize the model.

[0046] The image registration and merging process includes image registration, radiometric correction, and normalization. Image registration spatially registers images of different bands, aligning them in the same coordinate system to ensure a one-to-one correspondence between pixels of images of different bands. Radiometric correction performs radiometric correction on the images to eliminate radiometric differences introduced by sensor and atmospheric factors. Normalization standardizes pixel values ​​to the same range.

[0047] In step S1, Satellite sensors offer wide coverage and provide multispectral, high-resolution data, making them suitable for long-term monitoring of large areas. Unmanned aerial vehicle (UAV) sensors offer adjustable flight altitude and high flexibility, making them suitable for detailed monitoring of localized areas, and for high-precision monitoring and rapid response in specific regions. The visible light images reflect the color and reflectivity of the ocean surface and are used to detect ocean surface pollution and algae cover; the near-infrared images reflect the infrared reflectivity of vegetation and water bodies and are used to monitor the distribution and changes of marine vegetation and plankton; the short-wave infrared images have strong penetration and can provide information about water bodies and wetlands, and are used to monitor changes in ocean temperature, humidity and chemical composition.

[0048] In step S2, the feature training includes statistical analysis features, frequency domain analysis features, autocorrelation features, and histogram features; Statistical analysis characteristics include: Mean:

[0049] Where Mean represents the mean value of pixels in the image; K represents the total number of pixels in the image, z i This represents the pixel value of point i in the image; Variance:

[0050] Wherein, Variance represents the variance of the image pixel values; Standard Deviation:

[0051] Wherein, Standard Deviation represents the standard deviation of image pixel values; Kur:

[0052] Where Kur represents the kurtosis of the image; The kurtosis reflects the sharpness of the image brightness distribution; Frequency domain analysis features: Extracting frequency information from an image by performing a frequency domain transformation on the image; Fourier transform: transforms an image from the spatial domain to the frequency domain to analyze the frequency components of the image;

[0053] Where F(u,v) represents the image in the frequency domain; f(x,y) represents the image in the spatial domain; M represents the size of the image in the x-direction; N represents the size of the image in the y-direction; u represents the horizontal frequency component in the frequency domain; v represents the vertical frequency component in the frequency domain; and x and y represent the pixel coordinates in the spatial domain. Power spectral density E: describes the energy distribution of different frequency components in an image;

[0054] Where E(u,v) represents the power of each frequency component in the frequency domain; |F(u,v)| represents the amplitude of each frequency component in the frequency domain. High-frequency and low-frequency components of the image are extracted using a bandpass filter. High-frequency components: satisfying u≥u c , v≥v c u c It is half of the horizontal frequency range; v c It is half of the vertical frequency range; High-frequency component energy E high :

[0055] Total energy E total :

[0056] High-frequency component energy ratio (HF):

[0057] The high-frequency and low-frequency components of the image reflect the edge and texture features of the image as well as the overall smoothness of the image. Autocorrelation characteristics: The autocorrelation function is used to describe the correlation between pixels in an image;

[0058] Where K represents the total number of pixels in the image; R(Δx,Δy) represents the autocorrelation value between pixels at the displacement (Δx,Δy); and f(x+Δx,y+Δy) represents the image after displacement. Histogram characteristics: Histogram mean H Mean :

[0059] Among them, H Mean Let L represent the mean of the histogram; L represent the number of gray levels; p(t) represent the probability of gray level t. Histogram variance H var :

[0060] Among them, H var Indicates the variance of the histogram; Feature training is performed on the intelligent noise reduction model.

[0061] In step S2, noise category C k Including Gaussian noise, impulse noise, Poisson noise, and shot noise, the specific method for identification is as follows: In statistical features, kurtosis (Kur) is selected as the parameter:

[0062] Kur represents kurtosis; Indicates noise type C k The average value of the lower kurtosis; Indicates noise type C k The standard deviation below; Selecting the energy ratio of high-frequency components in the frequency domain characteristics:

[0063] Wherein, HF represents the energy ratio of the high-frequency components; Indicates noise type C k The average value of the energy ratio of the lower high-frequency components; Indicates noise type C k Standard deviation of the energy ratio of lower high-frequency components; Selecting the zero-delay autocorrelation value as a parameter from the autocorrelation features: the zero-delay autocorrelation value can be obtained by setting the values ​​of Δx and Δy to 0.

[0064] Where AC represents the zero-delay autocorrelation value; Indicates noise type C k The average value of zero-delay autocorrelation; Indicates noise type C k The standard deviation of the zero-delay autocorrelation value; Select the histogram mean as a parameter in the histogram features:

[0065] Among them, H var Indicates the variance of the histogram; Indicates noise type C k The average variance of the histogram below; Indicates noise type C k The standard deviation of the variance of the histogram below; Calculate the probability:

[0066]

[0067]

[0068]

[0069]

[0070] Among them, P(Kur), P(HF), P(AC) and P(H var P(C) represents the marginal probability of the feature; k () indicates noise type C k The prior probability, i.e., the probability that no noise feature was observed; P(Kur|C k ) indicates that in noise type C k Below, the probability density function of kurtosis; P(HF|C k ) indicates that in noise type C k Below, the probability density function of the energy ratio of high-frequency components; P(AC|C k ) indicates that in noise type C k Below, the probability density function of the zero-delay autocorrelation value; P(H var |C k ) indicates that in noise type C k Below, the probability density function of the histogram mean;

[0071] in, The noise type is then verified, including Gaussian noise, impulse noise, Poisson noise, and shot noise. The posterior probability of each noise type is calculated, and the noise type with the highest posterior probability is selected.

[0072] The Gaussian noise is reduced using wavelet transform, and the specific steps are as follows: S5-1. Decompose the image into different scales and directions through wavelet transform, and perform thresholding on the wavelet coefficients. S5-2. Perform wavelet transform on the image to obtain wavelet coefficients; perform soft thresholding and hard thresholding on the wavelet coefficients. S5-3. Perform inverse wavelet transform on the processed wavelet coefficients to reconstruct the image.

[0073] The impulse noise is reduced using adaptive median filtering, and the specific steps are as follows: S6-1. Initialize window size; S6-2. Calculate the median, maximum, and minimum values ​​of the pixels within the window; S6-3. Adjust the window size based on whether the pixel value is between the median and extreme values; S6-4. Replace the center pixel with the median of the adaptive window.

[0074] The method for processing Poisson noise is to use Anscombe transform to convert an image containing Poisson noise into a Gaussian noise-like image, then use wavelet transform for noise reduction, and finally perform inverse transform.

[0075] The Anscombe transform is specifically as follows:

[0076] Where I' represents a Gaussian noise image; I represents a Poisson noise image; The inverse transformation is:

[0077] Shot noise is reduced using gamma filtering.

[0078] The image registration and merging adopts a weighted average merging method, specifically:

[0079] in, This represents the pixel value at (x, y) in the merged image; , and represent the pixel values ​​at (x,y) of the visible light image, near-infrared image, and short-wave infrared image, respectively; w1, w2, and w3 represent the weights of the visible light image, near-infrared image, and short-wave infrared image, respectively. The weights are set according to the amount of information contained in the image, based on the experimental results, specifically w1=0.4, w2=0.3 and w3=0.3.

[0080] In step S4, the color correction and image enhancement specifically include: Color correction: Adjust the gain of each channel to make the image colors more balanced; select the white areas in the image and perform white balance correction to ensure that the white areas look more natural; use a reference image to perform color mapping so that the color distribution of the target image is consistent with the reference image; Image enhancement: Enhance image contrast to make bright and dark areas more distinct; perform unsharpening masking to enhance image edges and details, making the image clearer; convert to HSV color space and adjust color saturation and hue.

[0081] The system for intelligent denoising and enhancement of remote sensing data based on deep learning implements intelligent denoising and enhancement methods for remote sensing data based on deep learning. The system includes a data acquisition module, an intelligent denoising model module, an image registration and merging module, an image post-processing module, and a result display module. The data acquisition module is used to acquire remote sensing image data of the target area through satellite and UAV sensors; the intelligent noise reduction model module is used to process the acquired remote sensing images, remove various types of noise, including Gaussian noise, impulse noise, Poisson noise, and shot noise, and enhance image quality; the image registration and merging module is used to register and merge multispectral band images to generate a comprehensive image containing more information and details; the image post-processing module is used to further optimize image quality, enhance image details and contrast, and perform color correction and image enhancement; the result display module is used to display the processed image to the user, providing visualization tools and interfaces to support further analysis and operation of the image. The output of the data acquisition module is connected to the input of the intelligent noise reduction model module; the output of the intelligent noise reduction model module is connected to the input of the image registration and merging module; the output of the image registration and merging module is connected to the input of the image post-processing module; and the output of the image post-processing module is connected to the input of the result display module.

[0082] In an embodiment of the present invention, Step S1: Data Acquisition Data collection method: Satellite sensors: Using satellite sensors to cover the target ocean area to acquire multispectral high-resolution data, including visible light images, near-infrared images, and short-wave infrared images.

[0083] Drone sensors: Use drones for detailed monitoring of local areas to acquire high-precision visible light, near-infrared and short-wave infrared images.

[0084] Data type: Visible light images: used to detect marine surface pollution and algae cover.

[0085] Near-infrared images: used to monitor the distribution and changes of marine vegetation and plankton.

[0086] Shortwave infrared imagery: used to monitor changes in ocean temperature, humidity, and chemical composition.

[0087] Step S2: Build an intelligent noise reduction model Feature training: Statistical analysis features: Calculate the mean, variance, standard deviation, kurtosis, and skewness of the image.

[0088] Frequency domain analysis features: Extract the frequency components of the image through Fourier transform, and calculate the power spectral density and the energy ratio of the high-frequency components.

[0089] Autocorrelation feature: Calculates zero-delay autocorrelation values ​​to describe the correlation between pixels.

[0090] Histogram features: Calculate the mean and variance of the histogram to analyze the gray-level distribution of the image.

[0091] Intelligent noise reduction model: Gaussian noise: Wavelet denoising is used to remove Gaussian noise through wavelet transform and thresholding.

[0092] Impulse noise: Adaptive median filtering is used to adjust the filter window size according to the noise density to remove impulse noise.

[0093] Poisson noise: Use Anscombe transform to convert Poisson noise into Gaussian-like noise, and then use wavelet denoising.

[0094] Shot noise: Use gamma filtering for noise reduction.

[0095] Step S3: Image registration and merging Image registration: Spatial registration of images in different bands to align them in the same coordinate system and ensure one-to-one pixel correspondence.

[0096] Radiation correction: Eliminates radiation differences introduced by sensor and atmospheric factors.

[0097] Normalization: Standardizes pixel values ​​to the same range.

[0098] Weighted average merging: Based on experimentally determined weights, visible light, near-infrared and short-wave infrared images are merged.

[0099]

[0100] in, This represents the pixel value at (x, y) in the merged image; , and represents the pixel value at (x,y) of the visible light image, near-infrared image, and short-wave infrared image, respectively; w1, w2, and w3 represent the weights of the visible light image, near-infrared image, and short-wave infrared image, respectively; the weights are set according to the amount of information contained in the image based on the experimental conclusions, specifically w1=0.4, w2=0.3, and w3=0.3.

[0101] Step S4: Image Post-processing Color correction: White balance correction: Select the white area in the image for white balance correction.

[0102] Color mapping: Use a reference image for color mapping so that the color distribution of the target image is consistent with that of the reference image.

[0103] Image enhancement: Contrast Enhancement: Enhances the contrast between bright and dark areas of the image.

[0104] Unsharpened masking: Enhances the edges and details of the image.

[0105] Color saturation and hue adjustment: Convert to HSV color space and adjust color saturation and hue. Step S5: Visualizing Results and Providing Data Analysis Feedback Visualization: The processed images are displayed using image display software, making it easier for users to analyze and interpret them.

[0106] Data analysis feedback: The image data is analyzed in conjunction with specific application scenarios, and the analysis results are fed back to the intelligent noise reduction model to optimize the model.

[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0108] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent denoising and enhancement of remote sensing data based on deep learning, characterized by: The method includes the following steps: S1. Data acquisition: Collect remote sensing images of the target ocean area in different bands through satellite and UAV sensors; including visible light images, near-infrared images and short-wave infrared images; S2. Build an intelligent noise reduction model, train features through a deep learning network, and perform feature noise reduction processing on the acquired remote sensing image data; the feature training includes statistical analysis features, frequency domain analysis features, autocorrelation features, and histogram image features; Among them, noise category C k Including Gaussian noise, impulse noise, Poisson noise, and shot noise, the specific method for identification is as follows: In statistical features, kurtosis (Kur) is selected as the parameter: ； Kur represents kurtosis; Indicates noise type C k The average value of the lower kurtosis; Indicates noise type C k The standard deviation below; Selecting the energy ratio of high-frequency components in the frequency domain characteristics: ； Wherein, HF represents the energy ratio of the high-frequency components; Indicates noise type C k The average value of the energy ratio of the lower high-frequency components; Indicates noise type C k Standard deviation of the energy ratio of lower high-frequency components; Selecting the zero-delay autocorrelation value as a parameter from the autocorrelation features: the zero-delay autocorrelation value can be obtained by setting the values ​​of Δx and Δy to 0. ； Where AC represents the zero-delay autocorrelation value; Indicates noise type C k The average value of zero-delay autocorrelation; Indicates noise type C k The standard deviation of the zero-delay autocorrelation value; Select the histogram mean as a parameter in the histogram features: ； Among them, H var Indicates the variance of the histogram; Indicates noise type C k The average variance of the histogram below; Indicates noise type C k The standard deviation of the variance of the histogram below; Calculate the probability: ； Among them, P(Kur), P(HF), P(AC) and P(H var P(C) represents the marginal probability of the feature; k () indicates noise type C k The prior probability, i.e., the probability that no noise feature was observed; P(Kur|C k ) indicates that in noise type C k Below, the probability density function of kurtosis; P(HF|C k ) indicates that in noise type C k Below, the probability density function of the energy ratio of high-frequency components; P(AC|C k ) indicates that in noise type C k Below, the probability density function of the zero-delay autocorrelation value; P(H var |C k ) indicates that in noise type C k Below, the probability density function of the histogram mean; ； in, The noise type is indicated after verification, including Gaussian noise, impulse noise, Poisson noise, and shot noise; By calculating the posterior probability of each noise type, the noise type with the highest posterior probability is selected; Specifically, the Gaussian noise is denoised using wavelet transform; the impulse noise is denoised using adaptive median filtering; and the Poisson noise is processed by using Anscombe transform to convert the image containing Poisson noise into a Gaussian-like noise image, then using wavelet transform for denoising, and finally performing inverse transform. S3. Perform image registration and merging on images of different bands that have undergone feature denoising; S4. Perform post-processing on the images after registration and merging, including color correction and image enhancement; S5. Results visualization and data analysis feedback: The processed images are visualized using image display software; the image data is analyzed in conjunction with specific application scenarios, and the analysis results are fed back to the intelligent noise reduction model for optimization.

2. The method for intelligent noise reduction and enhancement of remote sensing data based on deep learning according to claim 1, characterized in that: The image registration and merging process includes image registration, radiometric correction, and normalization. Image registration spatially registers images of different bands, aligning them in the same coordinate system to ensure a one-to-one correspondence between pixels of images of different bands. Radiometric correction performs radiometric correction on the images to eliminate radiometric differences introduced by sensor and atmospheric factors. Normalization standardizes pixel values ​​to the same range.

3. The method for intelligent noise reduction and enhancement of remote sensing data based on deep learning according to claim 1, characterized in that: Statistical analysis characteristics include: Mean: ； Where Mean represents the mean value of pixels in the image; K represents the total number of pixels in the image, z i This represents the pixel value of point i in the image; Variance: ； Wherein, Variance represents the variance of the image pixel values; Standard Deviation: ； Wherein, Standard Deviation represents the standard deviation of image pixel values; Kur: ； Where Kur represents the kurtosis of the image; The kurtosis reflects the sharpness of the image brightness distribution; Frequency domain analysis features: Extracting frequency information from an image by performing a frequency domain transformation on the image; Fourier transform: transforms an image from the spatial domain to the frequency domain to analyze the frequency components of the image; ； Where F(u,v) represents the image in the frequency domain; f(x,y) represents the image in the spatial domain; M represents the size of the image in the x-direction; N represents the size of the image in the y-direction; u represents the horizontal frequency component in the frequency domain; v represents the vertical frequency component in the frequency domain; and x and y represent the pixel coordinates in the spatial domain. Power spectral density E: describes the energy distribution of different frequency components in an image; ； Where E(u,v) represents the power of each frequency component in the frequency domain; |F(u,v)| represents the amplitude of each frequency component in the frequency domain. High-frequency and low-frequency components of the image are extracted using a bandpass filter. High-frequency components: satisfying u≥u c , v≥v c u c It is half of the horizontal frequency range; v c It is half of the vertical frequency range; High-frequency component energy E high : ； Total energy E total : ； High-frequency component energy ratio (HF): ； The high-frequency and low-frequency components of the image reflect the edge and texture features of the image as well as the overall smoothness of the image. Autocorrelation characteristics: The autocorrelation function is used to describe the correlation between pixels in an image; ； Where K represents the total number of pixels in the image; R(Δx,Δy) represents the autocorrelation value between pixels at the displacement (Δx,Δy); and f(x+Δx,y+Δy) represents the image after displacement. Histogram characteristics: Histogram mean H Mean : ； Among them, H Mean Let L represent the mean of the histogram; L represent the number of gray levels; p(t) represent the probability of gray level t. Histogram variance H var : ； Among them, H var Indicates the variance of the histogram; Feature training is performed on the intelligent noise reduction model.

4. The method for intelligent noise reduction and enhancement of remote sensing data based on deep learning according to claim 3, characterized in that: The Gaussian noise is reduced using wavelet transform, and the specific steps are as follows: S5-1. Decompose the image into different scales and directions through wavelet transform, and perform thresholding on the wavelet coefficients. S5-2. Perform wavelet transform on the image to obtain wavelet coefficients; Soft and hard thresholding are applied to wavelet coefficients; S5-3. Perform inverse wavelet transform on the processed wavelet coefficients to reconstruct the image.

5. The method for intelligent noise reduction and enhancement of remote sensing data based on deep learning according to claim 3, characterized in that: The impulse noise is reduced using adaptive median filtering, and the specific steps are as follows: S6-1. Initialize window size; S6-2. Calculate the median, maximum, and minimum values ​​of the pixels within the window; S6-3. Adjust the window size based on whether the pixel value is between the median and extreme values; S6-4. Replace the center pixel with the median of the adaptive window.

6. The method for intelligent noise reduction and enhancement of remote sensing data based on deep learning according to claim 1, characterized in that: The image registration and merging adopts a weighted average merging method, specifically: ； in, This represents the pixel value at (x, y) in the merged image; , and represents the pixel value at (x,y) of the visible light image, near-infrared image, and short-wave infrared image, respectively; w1, w2, and w3 represent the weights of the visible light image, near-infrared image, and short-wave infrared image, respectively.

7. The method for intelligent noise reduction and enhancement of remote sensing data based on deep learning according to claim 1, characterized in that: In step S4, the color correction and image enhancement specifically include: The color correction is used to adjust the gain of each channel; white areas in the image are selected for white balance correction; a reference image is used for color mapping to make the color distribution of the target image consistent with that of the reference image; The image enhancement is used to enhance the contrast of the image; by using an unsharpening mask, the edges and details of the image are enhanced, and then the image is converted to the HSV color space to adjust the color saturation and hue.

8. A deep learning-based intelligent denoising and enhancement system for remote sensing data, executing the deep learning-based intelligent denoising and enhancement method for remote sensing data as described in any one of claims 1-7, characterized in that: The system includes a data acquisition module, an intelligent noise reduction model module, an image registration and merging module, an image post-processing module, and a result display module; The data acquisition module is used to acquire remote sensing image data of the target area through satellite and UAV sensors; the intelligent noise reduction model module is used to process the acquired remote sensing images, remove various types of noise, including Gaussian noise, impulse noise, Poisson noise, and shot noise, and enhance image quality; the image registration and merging module is used to register and merge multispectral band images to generate a comprehensive image containing more information and details; the image post-processing module is used to further optimize image quality, enhance image details and contrast, and perform color correction and image enhancement; the result display module is used to display the processed image to the user, providing visualization tools and interfaces to support further analysis and operation of the image. The output of the data acquisition module is connected to the input of the intelligent noise reduction model module; the output of the intelligent noise reduction model module is connected to the input of the image registration and merging module; the output of the image registration and merging module is connected to the input of the image post-processing module; and the output of the image post-processing module is connected to the input of the result display module.

Images