Method and system for processing and storing multivariate remote sensing data

Abstract

The invention discloses a method and a system for processing and storing multivariate remote sensing data. The method comprises the following steps: acquiring a multivariate remote sensing image and a demand database; inputting the multivariate remote sensing image into a preset image processing model to obtain multivariate interpretation information; and storing the multivariate interpretation information by adopting a reading probability model. According to the invention, the spectral image and the radar image are acquired by adopting multiple acquisition technologies, so that the richness and comprehensiveness of the image data are improved. And secondly, through mixed processing of the spectral image and the radar image, details of the image content are enhanced, and the performance and accuracy of image interpretation are also improved. Furthermore, the invention also covers a storage strategy based on a reading probability model, and realizes efficient storage of the multivariate interpretation information in combination with a compression processing technology, thereby not only reducing the storage cost, but also ensuring quick access of the data.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and image storage, and particularly relates to a method and system for processing and storing multi-source remote sensing data. Background Art

[0002] Remote sensing image technology plays a crucial role in many key fields such as agricultural monitoring, urban planning, natural resource management, and environmental protection. With the rapid development of remote sensing technology, the amount of remote sensing image data has increased explosively, driving the continuous rise in the demand for remote sensing image processing technology.

[0003] However, traditional remote sensing image processing methods often rely on a single acquisition device, which limits the richness and comprehensiveness of remote sensing image information, resulting in insufficient performance and accuracy in the image interpretation process. In addition, due to the large amount of remote sensing data, existing data storage strategies often overemphasize the reading speed while neglecting the reasonable planning of storage capacity, leading to high storage costs. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method and system for processing and storing multi-source remote sensing data. The technical problems to be solved by the present invention are achieved through the following technical solutions:

[0005] A method for processing and storing multi-source remote sensing data, comprising:

[0006] Obtaining multi-source remote sensing images and a demand database;

[0007] Inputting the multi-source remote sensing images into a preset image processing model to obtain multi-source interpretation information;

[0008] Storing the multi-source interpretation information using a reading probability model;

[0009] Wherein, the multi-source remote sensing images include spectral images and radar images; the multi-source interpretation information includes multi-source interpretation images and multi-source interpretation data.

[0010] In a specific embodiment, the inputting the multi-source remote sensing images into a preset image processing model to obtain multi-source interpretation information includes:

[0011] Obtaining high-definition spectral images based on the spectral images and the radar images;

[0012] Obtaining multi-source interpretation images based on the high-definition spectral images;

[0013] Obtaining multi-source interpretation data based on the multi-source interpretation images and the demand database.

[0014] In a specific embodiment, obtaining the high-definition spectral image based on the spectral image and the radar image includes:

[0015] Performing image alignment on the spectral image and the radar image to obtain a standard spectral image and a standard radar image, wherein the image alignment includes a feature matching algorithm and an image affine transformation;

[0016] Performing image filling on the standard spectral image and the standard radar image to obtain a high-definition spectral image.

[0017] In a specific embodiment, performing image filling on the standard spectral image and the standard radar image to obtain a high-definition spectral image includes:

[0018] According to the standard radar image, performing detail compensation on the standard spectral image to obtain a strong texture spectral image;

[0019] According to the standard radar image, performing color mapping and brightness mapping on the strong texture spectral image to obtain a corrected spectral image;

[0020] Performing image enhancement and edge smoothing on the corrected spectral image to obtain a high-definition spectral image.

[0021] In a specific embodiment, obtaining the multi-source interpretation image based on the high-definition spectral image includes:

[0022] Performing classification on the high-definition spectral image using a neural network to obtain an interest feature map;

[0023] Performing Unet convolutional neural network on the interest feature map and the high-definition spectral image to obtain a multi-source interpretation image;

[0024] Wherein, the classification neural network includes a plurality of convolutional kernels connected in sequence, 1 max pooling layer, a plurality of convolutional kernels, 1 average pooling layer, a plurality of convolutional kernels, 1 RELU activation function, and 2 fully connected layers;

[0025] The Unet convolutional neural network includes an encoder and a decoder. The encoder includes 3 encoding units connected in sequence, and each encoding unit includes a plurality of convolutional kernels, a normalization function, 1 RELU activation function, and 1 max pooling layer connected in sequence. The decoder includes 3 decoding units connected in sequence, and each decoding unit includes a plurality of convolutional kernels, a normalization function, 1 RELU activation function, and 1 upsampling process connected in sequence.

[0026] In a specific embodiment, obtaining the multi-source interpretation data based on the multi-source interpretation image and the demand database includes:

[0027] Obtaining demand feature data according to the demand database;

[0028] The multi - interpretation data is obtained by using a Probabilistic Neural Network (PNN) based on the requirement characteristic data and the multi - interpretation image.

[0029] In a specific embodiment, storing the multi - interpretation information by using a reading probability model includes:

[0030] The multi - interpretation image is split into a number of image units;

[0031] The multi - interpretation data is split into a number of data units;

[0032] The image units and the data units are processed by using a reading probability model and compression to obtain compression units;

[0033] The image units and the data units are subjected to hash calculation to obtain the unit information index of the corresponding compression units;

[0034] The compression units are processed by using a reading probability model and multi - level storage to obtain unit position indexes, so as to read the compression units according to the unit information indexes and the unit position indexes.

[0035] In a specific embodiment, processing the image units and the data units by using a reading probability model and compression to obtain compression units includes:

[0036] For image units with a reading probability greater than 80%, an image light compression method is used to obtain compression units;

[0037] For image units with a reading probability between 20% and 80%, an image medium compression method is used to obtain compression units;

[0038] For image units with a reading probability less than 20%, an image heavy compression method is used to obtain compression units;

[0039] For data units with a reading probability greater than 80%, a data light compression method is used to obtain compression units;

[0040] For data units with a reading probability between 20% and 80%, a data medium compression method is used to obtain compression units;

[0041] For data units with a reading probability less than 20%, a data heavy compression method is used to obtain compression units;

[0042] Among them,

[0043] The image light compression method includes pixel - adaptive prediction and single - code - table Golomb coding connected in sequence;

[0044] The image medium compression method includes multi - mode prediction and multi - code - table Golomb coding connected in sequence;

[0045] The described image recompression method includes multi-mode prediction, frequency-domain transformation, and multi-mode encoding connected in sequence;

[0046] The described data light compression method includes run-length encoding;

[0047] The described data medium compression method includes adaptive encoding of run-length encoding and bit-plane encoding;

[0048] The described data recompression method includes dictionary encoding, run-length encoding, and adaptive encoding of bit-plane encoding.

[0049] In a specific embodiment, obtaining the unit position index for the compression unit by using a read probability model and multi-level storage includes:

[0050] Using a read probability model for the compression unit to obtain the number of compression unit backups;

[0051] Using a read probability model for the compression unit to obtain the storage location of the compression unit;

[0052] Obtaining the unit position index based on the number of compression unit backups and the storage location of the compression unit.

[0053] The present invention also provides a processing and storage system for multi-source remote sensing images, including:

[0054] An acquisition unit for acquiring multi-source remote sensing images and a demand database;

[0055] A processing unit for inputting the multi-source remote sensing images into a preset generation neural network model to obtain multi-source interpretation information;

[0056] A storage unit for storing the multi-source interpretation information by using a read probability model;

[0057] Wherein, the multi-source remote sensing images include spectral images and radar images; the multi-source interpretation information includes multi-source interpretation images and multi-source interpretation data.

[0058] Advantages of the present invention:

[0059] A method and system for processing and storing multi-source remote sensing data according to the present invention first uses multiple acquisition technologies to obtain spectral images and radar images, thereby enhancing the richness and comprehensiveness of image data. Secondly, through the hybrid processing of spectral images and radar images, not only the details of the image content are enhanced, but also the performance and accuracy of image interpretation are improved. Further, the present invention also covers a storage strategy based on a read probability model, combined with compression processing technology, to achieve efficient storage of multi-source interpretation information, which not only reduces the storage cost but also ensures fast access to data.

[0060] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. Description of the Drawings

[0061] Figure 1 It is a flowchart of a method and system for processing and storing multi-source remote sensing data provided by an embodiment of the present invention;

[0062] Figure 2 It is a schematic diagram of high-definition spectral image synthesis in a method and system for processing and storing multi-source remote sensing data provided by an embodiment of the present invention;

[0063] Figure 3 It is a block diagram of modules in a method and system for processing and storing multi-source remote sensing data provided by an embodiment of the present invention. Detailed Embodiments

[0064] The present invention will be further described in detail below in conjunction with specific embodiments, but the embodiments of the present invention are not limited thereto.

[0065] Embodiment 1

[0066] In a specific embodiment, please refer to Figure 1 , Figure 1 which is a flowchart of a method and system for processing and storing multi-source remote sensing data, and the specific steps are as follows:

[0067] S1: Obtain multi-source remote sensing images and a requirements database, where the multi-source remote sensing images include spectral images and radar images; in a specific embodiment, traditional remote sensing image processing methods usually rely on a single acquisition device, which limits the diversity of acquired image data. To overcome this limitation, the present invention uses multiple acquisition devices for data acquisition to ensure that the acquired image data covers different spectral ranges. Therefore, in this embodiment, hyperspectral images or multispectral images are selected as representatives of spectral images. These images not only contain rich spectral information but also maintain high image clarity, providing the possibility for accurate ground object recognition and analysis.

[0068] At the same time, considering the advantages of synthetic aperture radar (SAR) images and interferometric synthetic aperture radar (InSAR) images in terms of anti-weather factor interference, and the fact that these radar images can provide a relatively high acquisition resolution, SAR and InSAR images are selected as representatives of radar images in this embodiment. These images can stably obtain surface information under various weather conditions, providing strong support for monitoring surface changes and terrain.

[0069] In terms of acquisition devices, the acquisition devices for spectral images include hyperspectral cameras and multispectral scanners, etc., which can capture a wide range of spectral bands from visible light to near-infrared or short-wave infrared. The acquisition devices for radar images include synthetic aperture radar systems, which can operate in different polarization modes and provide detailed surface structure and terrain information. Through these advanced acquisition devices, this embodiment can obtain high-quality spectral and radar images, laying a solid foundation for subsequent image processing and analysis.

[0070] S2: Input the multi-source remote sensing image into a preset image processing model to obtain multi-source interpretation information.

[0071] S21: Obtain a high-definition spectral image based on the spectral image and the radar image; in a specific embodiment, since the spectral image provides rich spectral information and can reflect the composition and structure of ground objects; while the radar image has all-weather and all-time imaging capabilities and has strong penetration for terrain and objects. Therefore, aligning these two images can integrate their respective advantages and improve the accuracy of information extraction and interpretation. Although the spectral image and the radar image provide different information about the same scene, due to differences in imaging mechanisms, wavelengths, resolutions, etc., these two images often need to be aligned before they can be compared or fused. The process of image alignment mainly includes two steps: a feature matching algorithm and an image affine transformation.

[0072] S211: Align the spectral image and the radar image to obtain a standard spectral image and a standard radar image, where the image alignment includes successively connecting a feature matching algorithm and an image affine transformation;

[0073] Feature matching algorithm:

[0074] In a specific embodiment, the specific steps are as follows:

[0075] 1. Use a feature detector (such as SIFT, SURF, ORB, etc.) to extract feature points in the spectral image and the radar image. These feature points are usually significant structures such as corners and edges in the image.

[0076] 2. Calculate the feature descriptors of each feature point to describe the local information of the feature point.

[0077] 3. Use a feature matching algorithm (such as BFMatcher) to find the most similar feature point pairs between the spectral image and the radar image according to the feature descriptors, which are used as the benchmark for image alignment.

[0078] Image affine transformation:

[0079] In a specific embodiment, the specific steps are as follows:

[0080] 1. Based on the matched feature point pairs, use the affine transformation algorithm to calculate the optimal transformation parameters of the affine transformation matrix for mapping the points in the radar image to the corresponding points in the spectral image, thereby achieving image alignment.

[0081] 2. Apply the transformation matrix to perform transformation processing on the radar image to align it with the spectral image, where the transformation processing includes operations such as translation, rotation, scaling, and skew.

[0082] S212: Please refer to Figure 2 , Figure 2 is a schematic diagram of the synthesis of high-definition spectral images in a method and system for processing and storing multi-source remote sensing data. The standard spectral image and the standard radar image are subjected to image filling to obtain a high-definition spectral image.

[0083] S2121: Although the spectral image provides rich spectral information, in some areas, especially those with less texture details, it may lack sufficient visual expressiveness. The standard radar image, with its high resolution and terrain penetration ability, can capture more detailed information. Therefore, combining the standard radar image to perform detail compensation on the standard spectral image aims to enhance the texture details of the spectral image and make it more visually impactful. Therefore, according to the standard radar image, detail compensation is performed on the standard spectral image to obtain a strong texture spectral image; after detail compensation, the texture details of the standard spectral image are significantly enhanced, forming a strong texture spectral image.

[0084] In a specific embodiment, the steps are as follows:

[0085] 1. Detail extraction: Extract the detail information of the high-frequency components from the standard radar image through a high-pass filter (such as a Laplacian filter, Sobel filter), where the detail information includes edges, contours, and microstructures, etc.

[0086] 2. Matching and fusion: First, match the extracted detail information with the standard spectral image to ensure the accurate correspondence of the detail information. Then, use image fusion technology to seamlessly integrate this detail information into the standard spectral image.

[0087] S2122: Although the strong texture spectral image has rich texture details, there may be inconsistencies in color and brightness. Although the standard radar image does not contain color information, its penetrability for terrain and objects helps to identify the reflection characteristics of different ground features. Therefore, color mapping and brightness mapping are performed on the strong texture spectral image in combination with the standard radar image, aiming to correct the color and brightness inconsistencies in the image and improve the accuracy and readability of the image. Therefore, according to the standard radar image, color mapping and brightness mapping are applied to the strong texture spectral image to obtain a corrected spectral image. After color mapping and brightness mapping, the color and brightness inconsistencies in the strong texture spectral image are corrected, forming a corrected spectral image. This image not only has rich texture details but also has more accurate and consistent color and brightness information.

[0088] In a specific embodiment, the steps are as follows:

[0089] 1. Color mapping: According to the reflection characteristics of different ground features in the standard radar image, a color mapping table is established. Then, the pixel values in the strong texture spectral image are matched with the mapping table to obtain the corrected color values.

[0090] 2. Brightness mapping: Similarly, according to the brightness information of the standard radar image, a brightness mapping table is established. The brightness values of the strong texture spectral image are adjusted to conform to the brightness distribution in the mapping table.

[0091] 3. Balance of comprehensive adjustment: In the process of comprehensive adjustment, it is necessary to balance the relationship between color, brightness, and texture to ensure the overall beauty and accuracy of the image.

[0092] Among them, when establishing the color mapping table and the brightness mapping table, it is necessary to fully consider the reflection characteristics and brightness distribution of different ground features. Moreover, it is necessary to ensure that the color and brightness mappings are accurately matched to avoid obvious color distortion or uneven brightness.

[0093] S2123: Although the corrected spectral image has rich texture details and accurate and consistent color and brightness information, there may still be problems such as insufficient contrast or rough edges in some areas. Therefore, image enhancement and edge smoothing techniques are used to further improve the contrast and clarity of the image, and at the same time smooth the edges to reduce edge burrs and artifacts. Therefore, image enhancement and edge smoothing are applied to the corrected spectral image to obtain a high-definition spectral image. After image enhancement and edge smoothing processing, the contrast and clarity of the corrected spectral image are significantly improved, and at the same time the edges are smoother, finally forming a high-definition spectral image. The high-definition spectral image not only has rich texture details, accurate and consistent color and brightness information, but also has high contrast and clarity.

[0094] In a specific embodiment, the steps are as follows:

[0095] Image enhancement: By using image enhancement techniques such as contrast enhancement and sharpening, the contrast and clarity of the corrected spectral image are improved, and the detailed information in the image is highlighted.

[0096] Edge smoothing: By using edge smoothing techniques such as smoothing filters, the edges in the image are smoothed, reducing edge burrs and artifacts, and improving the overall quality of the image.

[0097] S22: Obtain a multivariate interpretation image based on the high-definition spectral image.

[0098] S221: Although the high-definition spectral image provides rich spectral information and details, it is still a challenge to directly extract useful features from it for subsequent analysis. The classification neural network can automatically extract features useful for the classification task by learning the hierarchical features of the data. These features can better represent the key information in the image and provide a basis for subsequent multivariate interpretation. Therefore, the high-definition spectral image is processed by a classification neural network to obtain an interest feature map; after being processed by the classification neural network, the obtained interest feature map can more accurately reflect the key information in the high-definition spectral image.

[0099] In a specific embodiment, the steps are as follows:

[0100] 1. Take the high-definition spectral image as the input of the classification neural network.

[0101] 2. Convolution layer processing: 3 - 5 convolutional kernels perform convolution operations on the image to extract local features of the image. These convolutional kernels have different sizes and weights and can capture features of different scales and directions in the image.

[0102] 3. Max pooling layer: Downsample the convolved feature map through the max pooling layer to reduce the computational amount and retain important features.

[0103] 4. Repeat convolution and pooling: After 3 - 5 convolutional kernels and another average pooling layer, features are further extracted and compressed, while reducing the risk of noise and overfitting.

[0104] 5. Multiple convolutional kernels and RELU activation function: Extract features through 2 - 4 convolutional kernels and use the RELU activation function to increase the nonlinear ability of the network.

[0105] Fully connected layer: Map the extracted features to the classification space through two fully connected layers to obtain an interest feature map.

[0106] S222: The Unet convolutional neural network is a deep learning model commonly used in image segmentation tasks. It can utilize the structures of the encoder and decoder to effectively extract features in images and perform fine segmentation. By taking the interest feature map and the high-definition spectral image as inputs, Unet can further extract and integrate the information in the images to generate a more accurate multi-source interpretation image. Therefore, applying the Unet convolutional neural network to the interest feature map and the high-definition spectral image to obtain the multi-source interpretation image, the multi-source interpretation image can more accurately reflect different ground objects and features in the high-definition spectral image.

[0107] In a specific embodiment, the steps are as follows:

[0108] 1. Take the interest feature map and the high-definition spectral image as the inputs of the Unet convolutional neural network.

[0109] 2. In the encoder part, perform feature extraction on the input image through 3 encoding units. Each encoding unit includes 3 - 6 convolutional kernels, a normalization function, a RELU activation function, and a max pooling layer to gradually extract the deep features in the image.

[0110] 2. In the decoder part, perform upsampling and reconstruction on the features extracted by the encoder through 3 decoding units. Each decoding unit includes 3 - 6 convolutional kernels, a normalization function, a RELU activation function, and an upsampling process to gradually restore the resolution and details of the image.

[0111] Feature fusion: Between the encoder and the decoder, fuse the features extracted by the encoder and the features reconstructed by the decoder through skip connections to enhance the feature extraction ability of the network.

[0112] S23: Obtain multi-source interpretation data based on the multi-source interpretation image and the requirement database;

[0113] S231: The requirement database stores the specific requirements of users or application scenarios for multi-source interpretation data. By extracting the requirement feature data in the requirement database, the goals and key points of subsequent analysis and applications can be clarified, thus guiding the generation and processing of multi-source interpretation data. Therefore, obtain the requirement feature data based on the requirement database;

[0114] In a specific embodiment, the steps are as follows:

[0115] 1. Requirement database analysis: Analyze the requirement database to understand the specific requirements of users or application scenarios for multi-source interpretation data.

[0116] 2. Requirement feature extraction: According to the requirement analysis results, extract the requirement feature data in the requirement database. These feature data include ground object types, geomorphic features, vegetation coverage, etc.

[0117] 3. Feature data collation: Collate and summarize the extracted requirement feature data to provide a basis for subsequent analysis and application.

[0118] S232: The Probabilistic Neural Network (PNN) is a neural network model based on the Bayesian classifier. It can classify and predict based on the input feature data and training samples. By taking the requirement feature data and the multi-source interpretation image as inputs, the PNN can further analyze and process this data to generate multi-source interpretation data that meets the requirements of users or application scenarios. Therefore, multi-source interpretation data is obtained using the Probabilistic Neural Network (PNN) based on the requirement feature data and the multi-source interpretation image.

[0119] In a specific embodiment, the steps are as follows:

[0120] 1. Take the requirement feature data and the multi-source interpretation image as the inputs of the PNN.

[0121] 2. Feature extraction and mapping: In the PNN, the input data is subjected to feature extraction and mapping through the feature extraction layer, and the input data is converted into feature vectors suitable for classification and prediction.

[0122] 3. Calculate the probability distribution: Through the calculation layer, calculate the similarity between the feature vectors and the training samples, and calculate the probability distribution of the input feature vectors belonging to each category according to the similarity.

[0123] 4. Decision-making and output: Through the decision-making layer, make a decision based on the probability distribution and output the multi-source interpretation data.

[0124] S3: Store the multi-source interpretation information using a read probability model;

[0125] S31: Split the multi-source interpretation image to obtain image units;

[0126] S32: Split the multi-source interpretation data to obtain data units;

[0127] S33: Use a read probability model and compression processing on the image units and the data units to obtain compression units;

[0128] In a specific embodiment, the steps are as follows:

[0129] 1. Data reading and model update: Update a read probability model every day according to the reading quantities of the image units and the data units. This model can predict the likelihood of each data unit being read.

[0130] 2. Update of the probability model and identification of compression units: Based on the updated read probability model, identify the compression units that need to update the compression method according to the new read probability.

[0131] 3. Update of decompression and compression units: Decompress these identified compression units, and then execute the updated compression strategy according to the new read probability model.

[0132] 4. Re-compression: Re-compress these units according to the updated compression strategy to optimize the storage space and improve the read efficiency.

[0133] 5. Matching of data classification and compression algorithms: According to the read probability model, divide the image and data units into three different read probability levels: high, medium, and low. In this embodiment, the high read probability is set to 80%-100%, the medium read probability is 20%-80%, and the low read probability is 0%-20%.

[0134] For different read probability levels, adopt corresponding compression algorithms. Images or data with a high read probability use an algorithm with a low compression ratio to ensure fast decompression, while images or data with a low read probability can use a higher compression ratio to save storage space. In this way, it can be ensured that the data storage system can not only effectively utilize the storage space, but also quickly respond to high-frequency read requests, while optimizing the storage of infrequently accessed data. This dynamic compression management strategy helps to balance performance and storage costs and adapt to changing data access patterns.

[0135] The image light compression method includes sequentially connecting pixel adaptive prediction and single-codebook Golomb coding; wherein, for pixel adaptive prediction, according to the texture direction of the already encoded pixels around the current pixel, select the already encoded pixel closest to the current pixel in this texture direction as the reference pixel, and obtain the difference between the current pixel and the reference pixel as the prediction residual.

[0136] The image medium compression method includes sequentially connecting multi-mode prediction and multi-codebook Golomb coding; wherein, multi-mode prediction selects the prediction method with the smallest residual among pixel adaptive prediction and multiple fixed-direction predictions as the optimal prediction method.

[0137] The image heavy compression method includes sequentially connecting multi-mode prediction, frequency domain transformation, and multi-mode coding; wherein, multi-mode coding selects the coding method with the smallest coding bits among multi-codebook Golomb coding and fixed-length coding as the optimal coding method.

[0138] The data light compression method includes run-length encoding;

[0139] The data medium compression method includes run-length encoding and adaptive coding of bit-plane coding;

[0140] The data heavy compression method includes dictionary coding, run-length encoding, and adaptive coding of bit-plane coding.

[0141] S34: Calculate the unit information index of the corresponding compression unit for the image unit and the data unit using hash calculation;

[0142] S35: Obtain the unit location index for the compression unit using the read probability model and multi-level storage;

[0143] S351: By applying the read probability model, in this embodiment, the number of backups required for each compression unit is calculated. This decision is based on the importance and access frequency of the data to ensure the redundancy and reliability of critical data;

[0144] S352: Also based on the read probability model, in this embodiment, the best storage location is assigned to each compression unit. This decision takes into account the data popularity and the performance of the storage medium to achieve the best balance between cost and performance;

[0145] S353: Combine the number of backups of the compression unit and the storage location information to generate the location index for each unit. This index not only guides the physical storage of the data but also supports efficient data retrieval and recovery operations.

[0146] S36: Read the compression unit according to the unit information index and the unit location index.

[0147] In a specific embodiment, please refer to Figure 3 , Figure 3 is a block diagram of a module in a method and system for processing and storing multi-source remote sensing data. A multi-source remote sensing image processing and storage system, characterized in that:

[0148] Acquisition unit: Obtain multi-source remote sensing images and a demand database;

[0149] Processing unit: Input the multi-source remote sensing image into a preset generation neural network model to obtain multi-source interpretation information;

[0150] Storage unit: Store the multi-source interpretation information using the read probability model;

[0151] The multi-source remote sensing image includes a spectral image and a radar image; the multi-source interpretation information includes a multi-source interpretation image and multi-source interpretation data.

[0152] A method and system for processing and storing multi-source remote sensing data according to this embodiment first uses multiple acquisition techniques to obtain spectral images and radar images, thereby enhancing the richness and comprehensiveness of image data. Secondly, through the hybrid processing of spectral images and radar images, not only the details of the image content are enhanced, but also the performance and accuracy of image interpretation are improved. Further, the present invention also covers a storage strategy based on a read probability model, combined with compression processing technology, to achieve efficient storage of multi-source interpretation information, not only reducing the storage cost, but also ensuring fast access to data.

[0153] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0154] Although the present application has been described in conjunction with various embodiments herein, however, in the process of implementing the claimed present application, those skilled in the art can understand and achieve other variations of the disclosed embodiments by viewing the accompanying drawings, the disclosure content, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude a plurality of cases.

[0155] The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for processing and storing multi-source remote sensing data, characterized in that, Including: Obtain multi-source remote sensing images and a requirements database; Input the multi-source remote sensing images into a preset image processing model to obtain multi-source interpretation information; Store the multi-source interpretation information using a reading probability model; Wherein, the multi-source remote sensing images include spectral images and radar images; the multi-source interpretation information includes multi-source interpretation images and multi-source interpretation data.

2. The method for processing and storing multi-source remote sensing data according to claim 1, characterized in that The step of inputting the multi-source remote sensing images into a preset image processing model to obtain multi-source interpretation information includes: Obtain a high-definition spectral image based on the spectral image and the radar image; Obtain a multi-source interpretation image based on the high-definition spectral image; Obtain multi-source interpretation data based on the multi-source interpretation image and the requirements database.

3. A method for processing and storing multi-source remote sensing data according to claim 2, characterized in that, The step of obtaining a high-definition spectral image based on the spectral image and the radar image includes: Perform image alignment on the spectral image and the radar image to obtain a standard spectral image and a standard radar image, wherein the image alignment includes a feature matching algorithm and an image affine transformation; Perform image filling on the standard spectral image and the standard radar image to obtain a high-definition spectral image.

4. The method for processing and storing multi-source remote sensing data according to claim 3, characterized in that, The step of performing image filling on the standard spectral image and the standard radar image to obtain a high-definition spectral image includes: Perform detail compensation on the standard spectral image based on the standard radar image to obtain a strong texture spectral image; Perform color mapping and brightness mapping on the strong texture spectral image based on the standard radar image to obtain a corrected spectral image; Perform image enhancement and edge smoothing on the corrected spectral image to obtain a high-definition spectral image.

5. The method for processing and storing multi-source remote sensing data according to claim 2, characterized in that, The step of obtaining a multi-source interpretation image based on the high-definition spectral image includes: Use a classification neural network on the high-definition spectral image to obtain an interest feature map; Use a Unet convolutional neural network on the interest feature map and the high-definition spectral image to obtain a multi-source interpretation image; Wherein, the classification neural network includes a plurality of convolutional kernels connected in sequence, 1 max pooling layer, a plurality of convolutional kernels, 1 average pooling layer, a plurality of convolutional kernels, 1 RELU activation function, and 2 fully connected layers; The Unet convolutional neural network includes an encoder and a decoder, wherein the encoder includes 3 encoding units connected in sequence, each encoding unit includes a plurality of convolutional kernels, a normalization function, 1 RELU activation function, and 1 max pooling layer connected in sequence, and the decoder includes 3 decoding units connected in sequence, each decoding unit includes a plurality of convolutional kernels, a normalization function, 1 RELU activation function, and 1 upsampling process connected in sequence.

6. The method for processing and storing multi-source remote sensing data according to claim 2, wherein The step of obtaining multi-source interpretation data based on the multi-source interpretation image and the requirements database includes: Obtain requirement feature data based on the requirements database; Use a probability neural network PNN to obtain multi-source interpretation data based on the requirement feature data and the multi-source interpretation image.

7. A method for processing and storing multi-source remote sensing data according to claim 1, characterized in that, The step of storing the multi-source interpretation information using a reading probability model includes: Split the multi-source interpretation image to obtain a number of image units; Split the multi-source interpretation data to obtain a number of data units; Use a reading probability model and compression processing on the image units and the data units to obtain compression units; Hash calculation is performed on the image unit and the data unit to obtain the unit information index of the corresponding compression unit; The reading probability model and multi-level storage are used for the compression unit to obtain the unit position index, so as to read the compression unit according to the unit information index and the unit position index.

8. A method for processing and storing multi-source remote sensing data according to claim 7, characterized in that The step of using the reading probability model and compression processing for the image unit and the data unit to obtain the compression unit includes: For image units with a reading probability greater than 80%, an image light compression method is used to obtain the compression unit; For image units with a reading probability between 20% and 80%, an image medium compression method is used to obtain the compression unit; For image units with a reading probability less than 20%, an image heavy compression method is used to obtain the compression unit; For data units with a reading probability greater than 80%, a data light compression method is used to obtain the compression unit; For data units with a reading probability between 20% and 80%, a data medium compression method is used to obtain the compression unit; For data units with a reading probability less than 20%, a data heavy compression method is used to obtain the compression unit; Wherein, The image light compression method includes pixel adaptive prediction and single-codebook Golomb coding connected in sequence; The image medium compression method includes multi-mode prediction and multi-codebook Golomb coding connected in sequence; The image heavy compression method includes multi-mode prediction, frequency domain transformation and multi-mode coding connected in sequence; The data light compression method includes run-length encoding; The data medium compression method includes adaptive coding of run-length encoding and bit-plane coding; The data heavy compression method includes dictionary coding, run-length encoding and adaptive coding of bit-plane coding.

9. A method for processing and storing multi-source remote sensing data according to claim 8, characterized in that, The step of using the reading probability model and multi-level storage for the compression unit to obtain the unit position index includes: The reading probability model is used for the compression unit to obtain the number of compression unit backups; The reading probability model is used for the compression unit to obtain the storage location of the compression unit; The unit position index is obtained according to the number of compression unit backups and the storage location of the compression unit.

10. A processing and storage system for multi-source remote sensing images, characterized in that: An acquisition unit for acquiring multi-source remote sensing images and a demand database; A processing unit for inputting the multi-source remote sensing images into a preset generation neural network model to obtain multi-source interpretation information; A storage unit for storing the multi-source interpretation information using a reading probability model; Wherein, the multi-source remote sensing images include spectral images and radar images; the multi-source interpretation information includes multi-source interpretation images and multi-source interpretation data.

Images