Remote sensing image denoising method based on novel optimizer

Abstract

The invention discloses a remote sensing image denoising method based on a novel optimizer, which relates to the field of remote sensing image denoising, and comprises the following steps: collecting a remote sensing noise image, and dividing the remote sensing noise image into a training set and a test set; performing gradient optimization on the DudeNet denoising model by using a GAF algorithm, and training the DudeNet denoising model after gradient optimization by using the training set; and testing the trained DudeNet denoising model by using the test set, and evaluating by using an image quality measurement index to complete remote sensing image denoising based on the novel optimizer. The method solves the problems that an existing denoising method is poor in denoising effect and large in calculation overhead.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image denoising, and particularly to a remote sensing image denoising method based on a new optimizer. Background Art

[0002] Remote sensing images are widely used in fields such as geographic information monitoring, environmental monitoring, agricultural surveys, and urban planning. However, remote sensing images are often affected by noise during the acquisition process, especially in complex environments such as low light, clouds, sensor failures, or atmospheric interference, resulting in poor image quality and seriously affecting subsequent image analysis and interpretation. Therefore, the problem of remote sensing image denoising has become an important research direction in remote sensing image processing. Traditional remote sensing image denoising methods mainly include filtering algorithms, statistical models, etc., but these methods often have difficulty effectively removing noise while retaining image details. In recent years, deep learning-based denoising methods have become mainstream, but due to problems such as gradient disappearance and local optima during the training process of deep neural networks, the denoising effect is often poor or the computational cost is high. Therefore, how to design an efficient and accurate remote sensing image denoising model remains a current technical challenge. Summary of the Invention

[0003] Aiming at the above deficiencies in the prior art, the remote sensing image denoising method based on a new optimizer provided by the present invention solves the problems of poor denoising effect and high computational cost of existing denoising methods.

[0004] To achieve the above invention purpose, the technical solution adopted by the present invention is: a remote sensing image denoising method based on a new optimizer, including the following steps:

[0005] S1: Collect remote sensing noise images and divide the remote sensing noise images into a training set and a test set;

[0006] S2: Use the GAF algorithm to optimize the gradient of the DudeNet denoising model, and use the training set to train the DudeNet denoising model with the optimized gradient;

[0007] S3: Use the test set to test the trained DudeNet denoising model, and use image quality measurement indicators for evaluation to complete remote sensing image denoising based on a new optimizer.

[0008] Further, the S2 includes the following sub-steps:

[0009] S21: Use the scaling factor α and the sensitivity factor β in the GAF algorithm to adjust the gradient range of the GAF algorithm;

[0010] S22: Update the gradient of the GAF algorithm using the loss function, and replace the gradient of the DudeNet denoising model with the updated gradient of the GAF algorithm to obtain the DudeNet denoising model with optimized gradient;

[0011] S23: Use the training set to train the DudeNet denoising model with optimized gradient.

[0012] Further, in S21, the scaling factor α and the sensitivity factor β are set to {α = 0.1, β = 20} and {α = 0.2, β = 10}.

[0013] Further, in S22, the gradient of the GAF algorithm is updated using the loss function, and the formula is:

[0014]

[0015] where is the gradient of the random training process, ▽ is the gradient symbol, ω is the weight vector, i is the number of iterations, k is the number of data in the dataset, L(·) is the loss function, M(·) is the output of the model for each input sample in each iteration, X (·) is the input sample in the training batch, ω (·) is the weight vector, y (·) is the corresponding label of the input sample, g k is the gradient of the k-th iteration, μ m is the momentum coefficient, g k-1 is the gradient of the (k - 1)-th iteration, η is the learning rate, g' k is the gradient after momentum update applied to the GAF function to obtain the adjusted gradient, g'(·) is different types of GAF functions, ω k is the weight vector of the k-th iteration.

[0016] Further, in S22, the DudeNet denoising model includes a feature extraction module FEB, an image enhancement module EB, a compression module CB, and a reconstruction module RB.

[0017] Further, the feature extraction module FEB includes two parallel first feature extraction modules FEBnet1 and second feature extraction modules FEBnet2. The first feature extraction module FEBnet1 extracts global features and local features from the input noisy image through a sparse mechanism, and the second feature extraction module FEBnet2 is used to extract supplementary features from the input noisy image;

[0018] The first feature extraction module FEBnet1 adopts a sparse mechanism, including a convolutional layer, a batch normalization layer, and a ReLU activation layer;

[0019] The output FEB of the first feature extraction module FEBnet1 1 is expressed as:

[0020] FEB 1 = C(CBR 3 (S(CBR 1 (Y))))

[0021] where C(·) is a 3×3 convolution function, CBR 3 (·) is a module composed of three convolutional layers, a batch normalization layer, and a ReLU activation layer, S(·) is a sparse mechanism, CBR 1 (·) is a module composed of a convolutional layer, a batch normalization layer, and a ReLU activation layer, and Y is the input noisy image;

[0022] The second feature extraction module FEBnet2 includes a convolutional layer, a ReLU activation layer, and a convolutional layer connected in sequence;

[0023] The output FEB of the second feature extraction module FEBnet2 2 is expressed as:

[0024] FEB 2 = C 1 (CR 15 (Y))

[0025] where C 1 (·) is a 3×3 convolution function, CR 15 (·) is a functional module that sequentially executes 15 convolutional layers and ReLU activation layers.

[0026] Furthermore, the image enhancement module EB includes a first image enhancement module EB1 and a second image enhancement module EB2. The first image enhancement module EB1 is used to fuse the output features of the first feature extraction module FEBnet1 and the second feature extraction module FEBnet2, and process them through a batch normalization layer and a ReLU activation layer; the second image enhancement module EB2 is used to fuse the output features of the first image enhancement module EB1 and the input image, and compress the output features using a 1×1 convolution;

[0027] The output OE of the first image enhancement module EB1 1 is expressed as:

[0028] OE 1 = E(FEB 1 ,FEB 2 )

[0029] OE 1 = R(B(CON(C(CBR 3 (S(CBR1 (Y)))),CB(CR 15 (Y)))))

[0030] Among them, E(·) is the fusion operation, R(·) is the activation layer, B(·) is the batch normalization layer, CON(·) is the convolution function, and CB(·) is the compression module;

[0031] The output OE of the second image enhancement module EB2 2 is expressed as:

[0032] OE 2 = E(OCB 2 ,Y)

[0033] OCB 2 = C 1 (OE 1 )

[0034] Among them, OCB 2 represents the convolution of the output OE of the first image enhancement module EB1 1 to compress the data.

[0035] Furthermore, the compression module CB includes a first convolutional layer CB1, a second convolutional layer CB2, and a third convolutional layer CB3. The first convolutional layer CB1 is located at the end of the second feature extraction module FEBnet2. The second convolutional layer CB2 is located between the first image enhancement module EB1 and the second image enhancement module EB2. The third convolutional layer CB3 is located between the second image enhancement module EB2 and the reconstruction module RB;

[0036] The output of the compression module CB is expressed as:

[0037] OCB 3 = C 1 (OE 2 )

[0038] Among them, OCB 3 is the residual feature output by the third convolutional layer CB3.

[0039] Furthermore, the reconstruction module RB uses residual connection to map the compressed feature to the denoised image, and the output is expressed as:

[0040] X = Y - OCB 3

[0041] Among them, X is the reconstructed clear image.

[0042] Furthermore, the image quality measurement index in S3 is:

[0043]

[0044] Wherein, PSNR is an image quality measurement index, MAX is the maximum value of image intensity, MSE is the mean square error, p is the number of rows of image pixels, q is the number of columns of image pixels, r is the value of p, s is the value of q, I(·) is the pixel value of the original input image, and K(·) is the pixel value of the processed image.

[0045] The beneficial effects of the present invention are as follows:

[0046] (1) The GAF of the present invention modifies the gradient by applying a specific activation function. This function shrinks overly large gradients and amplifies tiny gradients, thereby preventing oscillations in sharp directions while accelerating convergence in flat directions. In terms of theoretical support: The optimizer used in this invention provides a theoretical proof that under certain conditions, GAF reduces the condition number of the optimization problem, which means higher stability and faster convergence.

[0047] (2) The design of the GAF of the present invention ensures that it avoids overly large or overly small gradient values, preventing the problems of gradient vanishing or explosion, especially in deep networks. The arctan-type GAF is emphasized as being particularly effective. In terms of theoretical support: The present invention provides a theorem showing that under specific conditions, the GAF output will have a larger magnitude than the small-gradient input, thereby preventing gradient vanishing. It also inherently limits the output of the gradient, helping to prevent gradient explosion.

[0048] (3) By modifying the gradient landscape, the GAF of the present invention allows the optimizer to more effectively move away from saddle points. This is achieved by reshaping the loss function, effectively making the saddle points less obvious. In terms of theoretical support: The technology of the present invention shows that under certain conditions, compared with the standard method, near the saddle point, the gradient descent method corrected by GAF converges to the minimum value faster.

[0049] (4) The present invention solves ill-conditioned problems and prevents gradient vanishing / explosion, and the GAF promotes faster convergence of the optimization algorithm. Theoretical support: The present invention theoretically proves that under certain assumptions, including the strong convexity and Lipschitz continuity conditions of the loss function, compared with SGD without GAF, the convergence rate of SGD with GAF is faster.

[0050] (5) The GAF of the present invention is easy to implement and integrate into existing deep learning frameworks (PyTorch and TensorFlow), and the computational cost of adding GAF is very small, only requiring element-wise activation on the gradient.

[0051] (6) The characteristics of GAF (amplifying small gradients and limiting large gradients) enable DudeNet to better adapt to various noise intensities and types. The technology of the present invention uses multiple types of GAF, such as arctan, tanh, and log types, which can be used to explore which GAF is more suitable for image denoising. Description of the Drawings

[0052] Figure 1 It is a flowchart of a remote sensing image denoising method based on a new optimizer.

[0053] Figure 2 It is a comparison chart of the differences in PSNR values of different optimizers in an image denoising model. Detailed Implementation Manner

[0054] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0055] As Figure 1 shown, a remote sensing image denoising method based on a new optimizer is characterized by including the following steps:

[0056] S1: Collect remote sensing noise images and divide the remote sensing noise images into a training set and a test set;

[0057] S2: Use the GAF algorithm to optimize the gradients of the DudeNet denoising model, and use the training set to train the DudeNet denoising model with optimized gradients;

[0058] S3: Use the test set to test the trained DudeNet denoising model and evaluate it using image quality measurement indicators to complete remote sensing image denoising based on a new optimizer.

[0059] In this embodiment, the training set uses 100 JPEG-compressed images, each image having a size of 512x512. The test set uses the online downloaded dataset CC, which contains 15 actual noise images of 512x512, and these images are taken by 3 digital devices: Canon 5D Mark III, Nikon D600, and Nikon D800.

[0060] The S2 includes the following sub-steps:

[0061] S21: Use the scaling factor α and the sensitivity factor β in the GAF algorithm to adjust the gradient range of the GAF algorithm;

[0062] S22: Use the loss function to update the gradients of the GAF algorithm, and replace the gradients of the DudeNet denoising model with the updated gradients of the GAF algorithm to obtain a DudeNet denoising model with optimized gradients;

[0063] S23: Train the denoising model of DudeNet with optimized gradients using the training set.

[0064] In S21, the scaling factor α and the sensitivity factor β are set to {α = 0.1, β = 20} and {α = 0.2, β = 10}.

[0065] In GAF, α and β are two important parameters used to control the shape and intensity of the gradient activation function. Together, they determine how GAF amplifies small gradients and limits large gradients.

[0066] Among them, α mainly controls the range of the output value of GAF. It determines the maximum amplitude by which the gradient value is scaled after being processed by GAF. When α is large, the output value range of GAF is also large, which means that after being processed by GAF, the gradient value may be amplified or reduced more, which is usually used to handle the situation where the gradient value is small. When α is small, the output value range of GAF is also small, which means that after being processed by GAF, the gradient value is amplified or reduced by a small amount. This is usually used to handle the situation where the gradient value is large.

[0067] β mainly controls the sensitivity of GAF when the gradient value is close to 0. It determines the steepness of the curve slope of GAF in the small gradient region. When β is large, the slope of GAF in the region where the gradient value is close to zero is steeper, which means that in the small gradient value region, GAF will significantly amplify the gradient, enabling these small gradients to be updated faster, thus avoiding gradient disappearance. When β is small, the slope of GAF in the region where the gradient value is close to zero is flatter, which means that in the small gradient value region, the amplification effect of GAF is weak.

[0068] Train on a neural network model without using GAF and record the maximum gradient value. Use visualization methods to draw the loss curve and compare it with the quadratic equation curve to determine. Increase the gradient for the regions in the loss curve that are flatter than the quadratic curve, and decrease the gradient for the regions that are steeper than the quadratic curve.

[0069] After multiple experiments, it is found that {α = 0.1, β = 20} and {α = 0.2, β = 10} are relatively stable, and initial attempts and fine-tuning can be carried out based on this.

[0070] In S22, the gradient of the GAF algorithm is updated using the loss function, and the formula is:

[0071]

[0072] Among them, is the gradient of the random training process, ▽ is the gradient symbol, ω is the weight vector, i is the number of iterations, k is the number of data in the dataset, L(·) is the loss function, M(·) is the output of the model for each input sample in each iteration, and X (·) is the input sample in the training batch, and ω (·) is the weight vector, and y (·) is the corresponding label of the input sample, and g k is the gradient of the k-th iteration, and μ m is the momentum coefficient, and g k-1 is the gradient of the (k - 1)-th iteration, η is the learning rate, and g' k is the gradient after applying the momentum update to the GAF function to obtain the adjusted gradient, g'(·) is different types of GAF functions, and ω k is the weight vector of the k-th iteration.

[0073] For different types of GAF, the corresponding formulas of g'(g k ) are different and are specifically as follows:

[0074] Arctan-type GAF: g'(g k ) = α arctan(βg k )

[0075] Tanh-type GAF: g'(g k ) = α tanh(βg k )

[0076] Log-type GAF: g'(g k ) = α(ln(ReLU(βg k ) + 1) - ln(ReLU(-βg k ) + 1))

[0077] After the above loop ends, the original gradient is replaced with the gradient activated by the GAF algorithm.

[0078] The DudeNet denoising model in S22 includes a feature extraction module FEB, an image enhancement module EB, a compression module CB, and a reconstruction module RB.

[0079] The feature extraction module FEB includes two parallel first feature extraction modules FEBnet1 and second feature extraction modules FEBnet2. The first feature extraction module FEBnet1 extracts global features and local features from the input noisy image through a sparse mechanism, and the second feature extraction module FEBnet2 is used to extract supplementary features from the input noisy image;

[0080] The first feature extraction module FEBnet1 adopts a sparse mechanism, including a convolutional layer, a batch normalization layer, and a ReLU activation layer;

[0081] The output FEB of the first feature extraction module FEBnet1 1 is expressed as:

[0082] FEB 1 = C(CBR 3 (S(CBR 1 (Y))))

[0083] where C(·) is a 3×3 convolution function, and CBR 3 (·) is a module composed of three convolutional layers, a batch normalization layer, and a ReLU activation layer, where CBR 3 can include dilated convolutions, S(·) is a sparse mechanism, and CBR 1 (·) is a module composed of a convolutional layer, a batch normalization layer, and a ReLU activation layer, and Y is the input noisy image;

[0084] The second feature extraction module FEBnet2 includes a convolutional layer, a ReLU activation layer, and a convolutional layer connected in sequence;

[0085] The output FEB of the second feature extraction module FEBnet2 2 is expressed as:

[0086] FEB 2 = C 1 (CR 15 (Y))

[0087] where C 1 (·) is a 3×3 convolution function, and CR 15 (·) is a functional module that sequentially executes 15 convolutional layers and ReLU activation layers.

[0088] The image enhancement module EB includes a first image enhancement module EB1 and a second image enhancement module EB2. The first image enhancement module EB1 is used to fuse the output features of the first feature extraction module FEBnet1 and the second feature extraction module FEBnet2, and process them through a batch normalization layer and a ReLU activation layer; the second image enhancement module EB2 is used to fuse the output features of the first image enhancement module EB1 and the input image, and compress the output features using 1×1 convolutions;

[0089] The output OE of the first image enhancement module EB1 1 is expressed as:

[0090] OE 1 = E(FEB 1,FEB 2 )

[0091] OE 1 = R(B(CON(C(CBR 3 (S(CBR 1 (Y)))),CB(CR 15 (Y)))))

[0092] Among them, E(·) is the fusion operation, R(·) is the activation layer, B(·) is the batch normalization layer, CON(·) is the convolution function, and CB(·) is the compression module;

[0093] The output OE of the second image enhancement module EB2 2 is expressed as:

[0094] OE 2 = E(OCB 2 , Y)

[0095] OCB 2 = C 1 (OE 1 )

[0096] Among them, OCB 2 represents the convolution of the output OE of the first image enhancement module EB1 1 to compress the data.

[0097] The compression module CB includes a first convolutional layer CB1, a second convolutional layer CB2, and a third convolutional layer CB3. The first convolutional layer CB1 is located at the end of the second feature extraction module FEBnet2, the second convolutional layer CB2 is located between the first image enhancement module EB1 and the second image enhancement module EB2, and the third convolutional layer CB3 is located between the second image enhancement module EB2 and the reconstruction module RB;

[0098] The output of the compression module CB is expressed as:

[0099] OCB 3 = C 1 (OE 2 )

[0100] Among them, OCB 3 is the residual feature output by the third convolutional layer CB3.

[0101] The reconstruction module RB uses residual connections to map the compressed features to the denoised image, and the output is expressed as:

[0102] X = Y - OCB 3

[0103] Among them, X is the reconstructed clear image.

[0104] The feature extraction module adopts a sparse mechanism, aiming to extract diverse features and reduce the depth of the network. The image enhancement module enhances the extracted features by fusing the features of two sub-networks, which is particularly useful for images contaminated by unknown types of noise. The compression module compresses the network to reduce the computational cost. The reconstruction module is used to finally reconstruct the clean image.

[0105] The optimized model is trained and the dataset images are denoised to obtain image pictures with better denoising effects, and the PSNR value is calculated for evaluation.

[0106] PSNR is a metric for measuring image quality, which is calculated by comparing the differences between the original image and the distorted image. Specifically, PSNR is calculated by comparing the pixel values of two images. If the two images are exactly the same, then the noise is zero and the PSNR is infinite. If the two images are very different, then the noise will be large and the PSNR will decrease accordingly. Therefore, the larger the PSNR, the better the image quality.

[0107] The image quality measurement index in S3 is as follows:

[0108]

[0109] Among them, PSNR is the image quality measurement index, MAX is the maximum value of the image intensity, MSE is the mean square error, p is the number of rows of the image pixels, q is the number of columns of the image pixels, r is the value of p, s is the value of q, I(·) is the pixel value of the original input image, and K(·) is the pixel value of the processed image.

[0110] In an embodiment of the present invention, as Figure 2 shown in Table 1, the PSNR differences of different optimizers in the image denoising model are compared.

[0111] Table 1 PSNR differences of different optimizers in the image denoising model

[0112]

[0113] Comparing the above experimental cases, the optimizer sgd-alt optimized by the GAF algorithm introduces a momentum term. When updating the parameters each time, it not only considers the current gradient but also combines the cumulative influence of the past gradients, jumps out of the saddle point during the training process, finds a better optimal solution, and effectively alleviates the problems of gradient explosion, gradient disappearance, saddle point, and ill-conditioned problems. In the denoising task, sgd-alt performs more stably and efficiently compared to the original adam optimizer, and has a better PSNR (peak signal-to-noise ratio) value for the same image denoising, and sgd-alt can achieve better denoising effects.

[0114] The GAF optimizer proposed by the present invention provides a unified solution to several common deep learning optimization challenges. This technology supports its claims through theoretical proofs and empirical evidence from experiments on various datasets and network architectures. The simplicity of implementation increases its attractiveness, indicating the potential for wide adoption.

[0115] Those of ordinary skill in the art will realize that the embodiments described herein are for helping the reader understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations without departing from the essence of the present invention based on these technical revelations disclosed by the present invention, and these deformations and combinations are still within the protection scope of the invention.

Claims

1. A remote sensing image denoising method based on a new optimizer, characterized in that: The following steps are involved: S1: Collecting remote sensing noise images and dividing the remote sensing noise images into a training set and a test set; S2: Use the GAF algorithm to perform gradient optimization on the DudeNet denoising model, and use the training set to train the gradient optimized DudeNet denoising model; S3: Use the test set to test the trained DudeNet denoising model and use image quality metrics to evaluate it, completing remote sensing image denoising based on the new optimizer.

2. A remote sensing image denoising method based on a novel optimizer according to claim 1, characterized in that: The S2 includes the following sub-steps: S21: Using the scaling factor in the GAF algorithm and sensitivity factor Adjust the gradient range of the GAF algorithm; S22: Use the loss function to update the gradient of the GAF algorithm, and replace the gradient of the DudeNet denoising model with the updated gradient of the GAF algorithm to obtain the gradient-optimized DudeNet denoising model; S23: Use the training set to train the gradient optimized DudeNet denoising model.

3. The remote sensing image denoising method based on the novel optimizer according to claim 2 is characterized in that: The S21 scaling factor and sensitivity factor Set to and .

4. The remote sensing image denoising method based on a novel optimizer according to claim 2, characterized in that: In S22, the loss function is used to update the gradient of the GAF algorithm, and the formula is: ； in, is the gradient of the stochastic training process, is the gradient symbol, is the weight vector, is the number of iterations, is the number of data sets, is the loss function, is the output of the model for each input sample in each iteration, is the input sample in the training batch, is the weight vector, is the corresponding label of the input sample, For the The gradient of the iteration, is the momentum coefficient, For the The gradient of the iteration, is the learning rate, The updated gradient for momentum is applied to the GAF function to obtain the adjusted gradient, For different types of GAF functions, For the The weight vector for the iteration.

5. The remote sensing image denoising method based on a novel optimizer according to claim 2, characterized in that: The DudeNet denoising model in S22 includes a feature extraction module FEB, an image enhancement module EB, a compression module CB and a reconstruction module RB.

6. A remote sensing image denoising method based on a novel optimizer according to claim 5, characterized in that: The feature extraction module FEB includes two parallel first feature extraction modules FEBnet1 and second feature extraction modules FEBnet2, wherein the first feature extraction module FEBnet1 extracts global features and local features from the input noise image through a sparse mechanism, and the second feature extraction module FEBnet2 is used to extract supplementary features from the input noise image; The first feature extraction module FEBnet1 adopts a sparse mechanism, including a convolution layer, a batch normalization layer and a ReLU activation layer; The output of the first feature extraction module FEBnet1 It is expressed as: ； in, for Convolution function, is a module consisting of three convolutional layers, a batch normalization layer, and a ReLU activation layer. is a sparse mechanism, is a module consisting of a convolutional layer, a batch normalization layer, and a ReLU activation layer. is the input noise image; The second feature extraction module FEBnet2 includes a convolutional layer, a ReLU activation layer and a convolutional layer connected in sequence; The output of the second feature extraction module FEBnet2 It is expressed as: ； in, for Convolution function, It is a functional module that executes 15 convolutional layers and ReLU activation layers in sequence.

7. A remote sensing image denoising method based on a novel optimizer according to claim 6, characterized in that: The image enhancement module EB includes a first image enhancement module EB1 and a second image enhancement module EB2, wherein the first image enhancement module EB1 is used to fuse the output features of the first feature extraction module FEBnet1 and the second feature extraction module FEBnet2, and process them through a batch normalization layer and a ReLU activation layer; the second image enhancement module EB2 is used to fuse the output features of the first image enhancement module EB1 and the input image, and compress the output features using a 1×1 convolution; The output of the first image enhancement module EB1 It is expressed as: ； ； in, represents the fusion operation, is the activation layer, is the batch normalization layer, is the convolution function, For compression module; The output of the second image enhancement module EB2 It is expressed as: ； ； in, Represents the output of the first image enhancement module EB1 Convolution is performed to compress the data.

8. The remote sensing image denoising method based on the novel optimizer according to claim 7 is characterized in that: The compression module CB includes a first convolutional layer CB1, a second convolutional layer CB2 and a third convolutional layer CB3, wherein the first convolutional layer CB1 is located at the end of the second feature extraction module FEBnet2, the second convolutional layer CB2 is located between the first image enhancement module EB1 and the second image enhancement module EB2, and the third convolutional layer CB3 is located between the second image enhancement module EB2 and the reconstruction module RB; The output of the compression module CB is represented as: ； in, It is the residual feature output by the third convolutional layer CB3.

9. A remote sensing image denoising method based on a novel optimizer according to claim 8, characterized in that: The reconstruction module RB uses residual connection to map the compressed features to the denoised image, and the output is expressed as: ； in, For the reconstructed clear image.

10. The remote sensing image denoising method based on a novel optimizer according to claim 1, characterized in that: The image quality measurement index in S3 is: ； ； in, is an indicator for measuring image quality. is the maximum value of the image intensity, is the mean square error, is the number of rows of image pixels, is the number of columns of image pixels, for The value of for The value of is the pixel value of the original input image, is the pixel value of the processed image.

Images