A remote sensing image super-resolution reconstruction method and system based on detail restoration

Abstract

The present invention relates to the field of image data processing technology, and in particular to a method and system for super-resolution reconstruction of remote sensing images based on detail recovery. The method comprises acquiring a remote sensing image; constructing a generative adversarial network model, including a generator and a discriminator; training the generator and the discriminator; extracting features from the remote sensing image using the generator to generate a reconstructed image; using the discriminator to discriminate between the reconstructed image and the high-resolution image of the remote sensing image to obtain a discrimination result; and setting a loss function according to the discrimination result to tune the generator and the discriminator. By adopting dynamic dense residual blocks and dynamic convolution technology, it can significantly improve the resolution and details of the image, thereby providing clearer and higher-quality remote sensing images.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to a remote sensing image super-resolution reconstruction method and system based on detail restoration. Background Art

[0002] As the demand for fine remote sensing increases in various industries, the application scenarios of high-resolution remote sensing images are becoming more and more extensive. Super-resolution reconstruction (SR) improves the quality and details of low-resolution images (LR) through algorithms. Currently, SR is receiving more and more attention and exploration in the field of remote sensing, including high-resolution map generation, multimodal data fusion, and target detection.

[0003] However, image super-resolution is a seriously ill-posed problem. Because the image degradation process is not deterministic, there is no precise solution. Given a LR, there are infinite possible solutions for the corresponding high-resolution image (HR). This makes SR a very challenging task. Therefore, studying an efficient and effective SR method to improve the quality of remote sensing images has great application effectiveness and practical significance.

[0004] Various efforts have been made to improve the perceived quality of images, such as CNN-based, Transformer-based architectures, and diffusion models. However, existing methods usually adopt local pixel smoothing strategies, such as the mean square error loss function. They cannot effectively capture the complex structure and non-local pixel dependencies of the image. Although good PSNR can be obtained, it is easy to cause the generated image to be too smooth and lack of realism and details. Therefore, it is necessary to develop more accurate and efficient super-resolution reconstruction methods to improve the resolution and quality of remote sensing images. Summary of the invention

[0005] In order to solve the above-mentioned problems, the present invention provides a remote sensing image super-resolution reconstruction method and system based on detail restoration.

[0006] In a first aspect, the present invention provides a remote sensing image super-resolution reconstruction method based on detail restoration, which adopts the following technical solution:

[0007] A remote sensing image super-resolution reconstruction method based on detail restoration, comprising:

[0008] Acquisition of remote sensing images;

[0009] Build a generative adversarial network model, including a generator and a discriminator;

[0010] Train the generator and the discriminator models separately;

[0011] The generator is used to extract features from remote sensing images and generate reconstructed images;

[0012] The discriminator is used to discriminate the high-resolution images of the reconstructed image and the remote sensing image to obtain a discrimination result;

[0013] The loss function is set according to the discrimination results to tune the generator and discriminator.

[0014] Furthermore, the construction of the generative adversarial network model includes constructing a generator, wherein the generator includes an OSRRDB, a convolutional layer, and an upsampling; wherein the generator includes a 3×3 convolutional layer, 23 OSRRDBs, and an upsampling module, each of which is formed by connecting an OSRDB with a self-attention mechanism. The dynamic residual dense block OSRRDB is composed of a dynamic convolution ODConv and a LeakyReLU.

[0015] Furthermore, the construction of the generative adversarial network model includes constructing a discriminator, and the discriminator adopts the architecture of VGG19. The VGG19 consists of 19 layers, including multiple 3×3 convolutional layers and average pooling layers. Its structure is divided into five groups of convolutional layers, and each group is followed by an average pooling layer.

[0016] Furthermore, the model training of the generator and the judge is performed separately, including introducing the PSNR indicator, training the PSNR-oriented generator, guiding the optimization process of the generator by calculating the PSNR value between the reconstructed image and the remote sensing image, and then using the trained generator as the initialization model to train the GAN model to achieve the final image quality optimization goal.

[0017] Furthermore, the model training of the generator and the judge separately also includes introducing the LPIPS indicator, and evaluating the similarity between the generated reconstructed image and the high-resolution image based on the LPIPS indicator to ensure that the generator can produce high-quality images that are closer to the real image.

[0018] Furthermore, the method of using the generator to extract features from the remote sensing image to generate a reconstructed image includes using the generator to extract features from the input image, then using OSRRDB to perform deep extraction and reconstruction of the input features, and finally further processing and amplifying the features through convolutional layers and upsampling layers to generate a reconstructed image.

[0019] Furthermore, the method uses a discriminator to discriminate the high-resolution images of the reconstructed image and the remote sensing image to obtain a discrimination result, including using the discriminator to receive an input image, processing the image through the convolution layer and pooling layer of the discriminator, converting it into a high-dimensional feature representation, and mapping the extracted features to the discrimination result through a fully connected layer.

[0020] In the second aspect, a remote sensing image super-resolution reconstruction system based on detail restoration includes:

[0021] The data acquisition module is configured to acquire remote sensing images;

[0022] The model building module is configured to build a generative adversarial network model, including a generator and a discriminator;

[0023] The model training module is configured to perform model training on the generator and the judger respectively;

[0024] The reconstruction module is configured to extract features from the remote sensing image using the generator to generate a reconstructed image;

[0025] The discrimination module is configured to discriminate the reconstructed image and the high-resolution image of the remote sensing image using the discriminator to obtain a discrimination result;

[0026] The tuning module is configured to tune the generator and the discriminator by setting a loss function according to the discrimination result.

[0027] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method for super-resolution reconstruction of remote sensing images based on detail recovery.

[0028] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the method for super-resolution reconstruction of remote sensing images based on detail recovery.

[0029] In summary, the present invention has the following beneficial technical effects:

[0030] The present invention provides a remote sensing image super-resolution reconstruction method based on detail recovery, and the present invention has significant advantages in the application field of remote sensing images. By adopting dynamic dense residual blocks and dynamic convolution technology, it can significantly improve the resolution and details of the image, thereby providing clearer and higher-quality remote sensing images. This improvement enhances the reliability of images in practical applications, such as environmental monitoring, urban planning, and disaster assessment, while also improving the accuracy of automated analysis. The improved image reconstruction technology effectively reduces noise and artifacts and improves data stability. In addition, higher-quality images can shorten the time for subsequent processing and analysis, save resources and costs, and thus improve overall work efficiency. In general, the present invention has broad practical value in tasks such as land object classification, target detection, and change detection, and supports more accurate decision-making and planning.

[0031] This method brings new technological breakthroughs to the field of remote sensing image processing, provides reliable technical support for related research and applications, and has good prospects for promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of a remote sensing image super-resolution reconstruction method based on detail recovery according to Example 1 of the present invention.

[0033] Figure 2 This is a diagram of the generator architecture of Example 1 of the present invention.

[0034] Figure 3 It is a schematic diagram of the discriminator structure of embodiment 1 of the present invention.

[0035] Figure 4 It is the original high-resolution image of Example 1 of the present invention.

[0036] Figure 5 It is another original high-resolution image of Example 1 of the present invention.

[0037] Figure 6 It is a low-resolution image with a Gaussian degradation kernel size of 0.6 according to Example 1 of the present invention.

[0038] Figure 7 It is a low-resolution image with a Gaussian degradation kernel size of 1.2 according to Example 1 of the present invention.

[0039] Figure 8 It is a low-resolution image with a Gaussian degradation kernel size of 1.8 according to Example 1 of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0041] Example 1

[0042] Reference Figure 1 , a remote sensing image super-resolution reconstruction method based on detail restoration in this embodiment includes:

[0043] Acquisition of remote sensing images;

[0044] Build a generative adversarial network model, including a generator and a discriminator;

[0045] Train the generator and the discriminator models separately;

[0046] The generator is used to extract features from remote sensing images and generate reconstructed images;

[0047] The discriminator is used to discriminate the high-resolution images of the reconstructed image and the remote sensing image to obtain a discrimination result;

[0048] The loss function is set according to the discrimination results to tune the generator and discriminator.

[0049] Specifically:

[0050] S1. Construct the generator dynamic convolution and self-attention residual dense block network OSRRDBNet, referred to as the generator. The generator can be roughly divided into three substructures: OSRRDB, convolution layer and upsampling. First, the generator extracts features from the input image, and then uses OSRRDB to perform deep extraction and reconstruction of the input features. Finally, the convolution layer and upsampling layer further process and amplify the image to generate a reconstructed image. This design greatly improves the accuracy of super-resolution reconstruction of remote sensing images. During the training process, the generator attempts to map the input low-resolution image to a space that is more similar to the real high-resolution image to generate a reconstructed image;

[0051] This process can be expressed as:

[0052]

[0053] Where Y is the reconstructed high-resolution image and X is the input low-resolution image. Represents a 3*3 convolution operation, Indicates that the processing operation is performed through the OSRRDB block. is an upsampling operation.

[0054] like Figure 2 As shown, the key component of the generator network OSRRDBNet of this embodiment is the dynamic dense residual block OSRRDB. Through the residual connection, the model can more effectively learn and reconstruct the details and features of the input image.

[0055] Each OSRRDB is composed of an OSRDB connected with a simple self-attention mechanism. OSRDB mainly consists of dynamic convolution, and LeakyReLU. Dynamic convolution and activation units are combined through residual connections, so that each dynamic dense residual block can fully utilize the input and output of the previous layer. The design of OSRDB facilitates feature reuse and information transfer, thereby enhancing the ability of feature extraction. Compared with the RRDB architecture containing 3×3 convolution and activation layers, this design can effectively capture long-term dependencies and important features in the image.

[0056] Dynamic Convolution (ODConv): Dynamic convolution is one of the key features of OSRRDB. Compared with traditional 3*3 convolution, dynamic convolution can dynamically adjust the shape and parameters of the convolution kernel according to the characteristics of the input data. This enables the network to better adapt to features of different types and scales, thereby more effectively capturing details and structures in the image.

[0057] Simple self-attention mechanism: The self-attention mechanism used in OSRRDB is a lightweight attention mechanism used to enhance the model's global perception. By introducing self-attention in each OSRRDB, the network can better understand the dependencies between various parts of the image, which helps improve the accuracy and quality of image reconstruction.

[0058] LeakyReLU activation function: Compared to traditional ReLU, LeakyReLU introduces a small negative slope when processing the activation function. This makes the network more stable during training and can handle a wider range of input distributions.

[0059] The OSRRDB of the present invention integrates residual connections, dense blocks and self-attention mechanisms through its unique structural design, so that the model can perform well in the task of super-resolution reconstruction of optical remote sensing images. It can not only effectively capture long-term dependencies, but also extract and retain important features in the image. This significantly improves the quality and visual effect of the final reconstructed image.

[0060] S2. For the discriminator, the present invention designs an architecture that accurately extracts VGG19. An average pooling layer is used, which can prompt the generator to generate images that are more similar to HR. Aiming at the two indicators of PSNR and LPIPS, the generator and discriminator are iteratively trained and updated until the reconstructed image converges to be similar to the original high-resolution image;

[0061] Among them, Figure 3 As shown in the figure, VGG19 is a deep convolutional neural network consisting of 19 layers, including multiple 3×3 convolutional layers and average pooling layers. Its structure is divided into five groups of convolutional layers, each followed by an average pooling layer. This architecture is designed to extract image features by using small convolution kernels and multiple layers of depth. The shallow convolutional layers and pooling layers of the discriminator network are used to extract low-level and high-level features of the image. The convolutional layers extract features at different levels, such as edges, textures, and complex patterns, while the pooling layers are used to reduce the spatial dimensions of the feature maps and reduce the amount of calculation. These layers can help extract features of generated images and real images to evaluate the similarity between generated images and real images. The deep convolutional layers and fully connected layers of the network are mainly used to provide high-level feature descriptions of the image. These features are crucial for evaluating the quality of generated images.

[0062] S3. The model training of the present invention is divided into two parts. First, the PSNR-oriented generator is trained. The goal of this stage is to improve the overall quality of the generator when generating high-resolution images. Taking the high-resolution image as a reference, the optimization process of the generator is guided by calculating the PSNR value between the generated image and the real image. Then, the GAN model is trained with the trained generator as the initialization model to achieve the final image quality optimization goal, wherein the GAN model is a remote sensing image super-resolution reconstruction model based on detail recovery. In this stage, the generator and the discriminator are optimized through alternating training. The generator generates images, while the discriminator evaluates the authenticity of the generated images and provides feedback. Through adversarial training, not only the PSNR value of the generated image is improved, but also the visual quality of the image is enhanced, making the generated high-resolution image more outstanding in details, texture and realism;

[0063] in,

[0064] The goal of the generator G is to use the generated samples to deceive the discriminator. Its objective function is as follows:

[0065]

[0066] The goal of the discriminator D is to identify the authenticity of the input sample, and its objective function is as follows:

[0067]

[0068] Therefore, the overall objective function can be summarized as:

[0069]

[0070] Among them, G(z) is the reconstructed image and D(x) represents the probability that the result is the real image.

[0071] Among them, the present invention uses high-resolution images and corresponding low-resolution images as training data. The real image in the validation set will be used as a reference image to calculate the PSNR value between the generated image and the real image. During the training process, we use a loss function to guide fine-tuning training. The higher the PSNR value, the higher the similarity between the generated image and the real image. During the training process, the PSNR value is improved by optimizing the generator, thereby improving the quality of the image. Then, the generator trained in the first stage is used as the initialization model to ensure that the generator has a high generation quality. In the second stage, a complete reconstruction model is trained. The generator and the discriminator are trained alternately. The generator tries to generate real images that are more difficult to distinguish, while the discriminator continuously improves the ability to distinguish between generated images and real images. The two promote each other in the confrontation process, and finally achieve the purpose of generating higher quality images. The parameters of the generator and the discriminator are adjusted by the Adam optimizer, so that the images generated by the generator are closer and closer to the real images, and the discriminator continuously improves the ability to distinguish between generated images and real images. After adversarial training, the images generated by the generator are not only improved in PSNR value, but also significantly improved in visual quality. The details, texture and overall realism of the image will be enhanced.

[0072] S4. During the training process, in order to more accurately evaluate the visual quality of the generated images, the present invention introduces the LPIPS indicator, which can better simulate human perception. By monitoring the changes in the PSNR and LPIPS indicators of the model on the validation set, the similarity between the generated SR images and the HR images is regularly evaluated to ensure that the generator can produce high-quality images that are closer to the real images. Through this training optimization strategy, not only the sensitivity of traditional indicators to numerical errors is taken into account, but also a deeper understanding of the perceived quality of the image is added. It enables us to more comprehensively evaluate the performance of the generator and further optimize the training process to obtain more realistic super-resolution images.

[0073] in,

[0074] PSNR measures the absolute difference between the generated image and the real image, and the formula is as follows:

[0075]

[0076] Where R is the dynamic range of the image (e.g., for an 8-bit image, R=255) and MSE is the mean squared error.

[0077] LPIPS measures the perceptual differences between images. Its calculation involves feeding an image into a pre-trained deep network and computing the differences in their feature maps:

[0078]

[0079] Where L is the number of network layers, is the feature map extracted at the lth layer, and is the Euclidean distance.

[0080] In the process of training the complete reconstruction model, we use an iterative optimization method to alternately update the generator and the discriminator. The discriminator compares the generated SR image with the real HR image and feeds back information about the image quality to the generator, helping the generator to continuously optimize the generation process. However, the traditional evaluation indicators PSNR and SSIM used in previous studies may have difficulty in evaluating the perceptual image quality in the generated images.

[0081] In order to more accurately evaluate the visual quality of generated images, we introduced the LPIPS metric, which can better simulate human perception. By monitoring the changes in the PSNR and LPIPS metrics of the model on the validation set, the similarity between the generated SR images and HR images is regularly evaluated to ensure that the generator can produce high-quality images that are closer to real images.

[0082] Through this training optimization strategy, we not only consider the sensitivity of traditional indicators to numerical errors, but also incorporate a deeper understanding of the perceptual quality of images. This allows us to more comprehensively evaluate the performance of the generator and further optimize the training process to obtain more realistic super-resolution images.

[0083] Experimental results:

[0084] The remote sensing image reconstruction method of the present application is used to reconstruct the remote sensing images in the AID dataset. The reconstructed low-resolution images are obtained by bicubic downsampling. The reconstruction results are as follows: Figure 4 , Figure 5 shown.

[0085] in, Figure 4 and Figure 5 (a) is the original high-resolution image, (b) is the locally enlarged image of the original high-resolution image, (c) is the image obtained by bicubic interpolation, (d) is the image obtained by the generative adversarial network model, (e) is the image obtained by the enhanced generative adversarial network model, (f) is the image obtained by the blind super-resolution reconstruction model, (g) is the image obtained by the lightweight image super-resolution model, (h) is the image obtained based on the diffusion probability model, and (i) is the image obtained by the remote sensing image reconstruction method of the present application.

[0086] In addition, the remote sensing image reconstruction method of the present application is also used to reconstruct the WHU-RS19 dataset. The low-resolution images required for reconstruction are obtained by processing with Gaussian degradation kernels of different scales, as shown in the following figure. Figure 6 , Figure 7 , Figure 8 shown. Figure 6The Gaussian degenerate kernel size is 0.6, Figure 7 The Gaussian degenerate kernel size is 1.2, Figure 8 The Gaussian degenerate kernel size is 1.8.

[0087] Among them, the first row is the original or reconstructed image of each model, and the second row is the corresponding locally enlarged image. (a) is the original high-resolution image, (b) is the locally enlarged image of the image obtained by bicubic interpolation, (c) is the locally enlarged image of the image obtained by the generative adversarial network model, (d) is the locally enlarged image of the image obtained by the enhanced generative adversarial network model, (e) is the locally enlarged image of the image obtained by the blind super-resolution reconstruction model, (f) is the image obtained by the lightweight image super-resolution model, (g) is the image obtained based on the diffusion probability model, and (h) is the locally enlarged image of the image obtained by the remote sensing image reconstruction method of the present application.

[0088] The image obtained by bicubic interpolation is obtained by upscaling the low-resolution image to a high resolution without adding extra details or textures;

[0089] The image is obtained through the generative adversarial network model, which is a classic image super-resolution method. The mapping relationship from low resolution to high resolution is learned through the generative adversarial network model, but it may cause some image distortion or blur;

[0090] The image is obtained through the enhanced generative adversarial network model, and the reconstruction residual block and perceptual loss function are introduced, which helps to improve the quality of the generated image and better preserve the image details and texture;

[0091] The images obtained by the blind super-resolution reconstruction model propose a more complex but practical image degradation model, which aims to better simulate the degradation of images in the real world.

[0092] The images obtained by the lightweight image super-resolution model can effectively handle image super-resolution tasks while ensuring computational efficiency through convolution kernel and feature shuffling technology.

[0093] Based on the image obtained by the diffusion probability model, a remote sensing image super-resolution method based on the diffusion probability model is proposed, which aims to solve the over-smoothing problem of traditional convolutional networks and the artifact problem of generative adversarial networks.

[0094] Depend on Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8It can be seen that compared with other methods, the remote sensing image reconstruction method of the present application can more effectively simulate the degradation of real images and adaptively fuse information of different resolutions during the reconstruction process, thereby improving the clarity and accuracy of the reconstructed image.

[0095] Example 2

[0096] This embodiment provides a remote sensing image super-resolution reconstruction system based on detail restoration, including:

[0097] The data acquisition module is configured to acquire remote sensing images;

[0098] The model building module is configured to build a generative adversarial network model, including a generator and a discriminator;

[0099] The model training module is configured to perform model training on the generator and the judger respectively;

[0100] The reconstruction module is configured to extract features from the remote sensing image using the generator to generate a reconstructed image;

[0101] The discrimination module is configured to discriminate the reconstructed image and the high-resolution image of the remote sensing image using the discriminator to obtain a discrimination result;

[0102] The tuning module is configured to tune the generator and the discriminator by setting a loss function according to the discrimination result.

[0103] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a remote sensing image super-resolution reconstruction method based on detail recovery.

[0104] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the remote sensing image super-resolution reconstruction method based on detail recovery.

[0105] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A remote sensing image super-resolution reconstruction method based on detail restoration, characterized in that: include: Acquisition of remote sensing images; Build a generative adversarial network model, including a generator and a discriminator; Train the generator and the discriminator models separately; The generator is used to extract features from remote sensing images and generate reconstructed images; The discriminator is used to discriminate the high-resolution images of the reconstructed image and the remote sensing image to obtain a discrimination result; According to the discrimination results, the loss function is set to optimize the generator and discriminator; The construction of the generative adversarial network model includes constructing a generator, wherein the generator includes an OSRRDB, a convolutional layer and an upsampling; wherein the generator includes a 3×3 convolutional layer, 23 OSRRDBs and an upsampling module, each of which is formed by connecting an OSRDB with a self-attention mechanism, and the dynamic residual dense block OSRRDB is composed of a dynamic convolution ODConv and a LeakyReLU; the dynamic convolution ODConv is a key feature of OSRRDB, and is used to dynamically adjust the shape and parameters of the convolution kernel according to the features of the input data, so that the network adapts to features of different types and scales, thereby capturing details and structures in the image; The constructing of the generative adversarial network model includes constructing a discriminator, the discriminator adopts the architecture of VGG19, the VGG19 is composed of 19 layers, including multiple 3×3 convolutional layers and average pooling layers, and its structure is divided into five groups of convolutional layers, each group is followed by an average pooling layer; the model training of the generator and the discriminator respectively includes introducing the PSNR indicator, training the PSNR-oriented generator, guiding the optimization process of the generator by calculating the PSNR value between the reconstructed image and the remote sensing image, and then using the trained generator as the initialization model to train the GAN model to achieve the final image quality optimization goal; The model training of the generator and the judger respectively also includes introducing the LPIPS indicator, and evaluating the similarity between the generated reconstructed image and the high-resolution image based on the LPIPS indicator to ensure that the generator can generate high-quality images; Through training optimization strategies, we not only consider the sensitivity of traditional indicators to numerical errors, but also incorporate a deep understanding of image perceptual quality, which enables us to comprehensively evaluate the performance of the generator and further optimize the training process to obtain more realistic super-resolution images. The method comprises extracting features of the remote sensing image using the generator to generate a reconstructed image, comprising extracting features of the input image using the generator, extracting and reconstructing the input features in depth using OSRRDB, and further processing and amplifying the input features through a convolution layer and an upsampling layer to generate a reconstructed image. The method uses a discriminator to discriminate high-resolution images of reconstructed images and remote sensing images to obtain a discrimination result, including using the discriminator to receive an input image, processing the image through a convolution layer and a pooling layer of the discriminator, converting it into a high-dimensional feature representation, and mapping the extracted features to the discrimination result through a fully connected layer.

2. A remote sensing image super-resolution reconstruction system based on detail recovery, executing the method according to claim 1, characterized in that: include: The data acquisition module is configured to acquire remote sensing images; The model building module is configured to build a generative adversarial network model, including a generator and a discriminator; The model training module is configured to perform model training on the generator and the judger respectively; The reconstruction module is configured to extract features from the remote sensing image using the generator to generate a reconstructed image; The discrimination module is configured to discriminate the reconstructed image and the high-resolution image of the remote sensing image using the discriminator to obtain a discrimination result; The tuning module is configured to tune the generator and the discriminator by setting the loss function according to the discrimination result.

3. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method according to claim 1 .

4. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method as claimed in claim 1 .

Images