Intelligent ground feature information extraction method and system based on remote sensing image big data

Abstract

The invention belongs to the technical field of remote sensing images, particularly relates to an intelligent ground feature information extraction method and system based on remote sensing image big data, electronic equipment and a computer readable storage medium, and aims to solve the problems of poor accuracy and robustness during ground feature information extraction and category recognition in the prior art. The method comprises the following steps: acquiring remote sensing image data of a ground object to be subjected to category recognition as input data; preprocessing the input data to obtain preprocessed data; extracting features of the ground object based on the preprocessed data, and fusing the extracted features to obtain fused features; inputting the fusion features into a trained ground object classification model to obtain a category recognition result of the ground object; wherein the ground feature classification model is constructed based on a deep learning model. According to the invention, the accuracy and robustness of ground feature information extraction and category identification are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing image technology, and specifically relates to an intelligent ground object information extraction method, system, electronic equipment and computer-readable storage medium based on remote sensing image big data. Background Art

[0002] With the rapid development of satellite technology and drone technology, it is becoming easier and easier to obtain high-quality and high-frequency remote sensing images. These images not only have higher resolution and wider coverage, but also can provide rich information such as multispectral, hyperspectral and even thermal infrared. However, most traditional land object classification methods rely on artificially designed features such as texture, shape and spectral features, which are incapable of dealing with complex natural scenes and diverse land object types. Due to the complexity and diversity of land object types in nature, as well as changes in environmental conditions (such as weather and light), it is difficult for traditional methods to adaptively adjust feature extraction strategies, resulting in limited classification accuracy.

[0003] In recent years, deep learning technology has achieved remarkable results in image classification tasks due to its powerful nonlinear mapping capabilities and ability to learn complex patterns. Especially in the field of computer vision, deep learning models can automatically learn effective feature representations from data without human intervention. This feature makes them particularly suitable for processing remote sensing images because they can capture subtle differences between objects and improve the accuracy and robustness of category recognition.

[0004] Despite this, existing models still face a series of challenges when processing remote sensing images. First, insufficient fusion of multi-source data is an important issue. The data provided by different sensors have different spatial resolutions, spectral responses, and temporal resolutions. How to effectively integrate these heterogeneous data to fully utilize their respective advantages remains an urgent problem to be solved. Secondly, the insufficient utilization of spatiotemporal features is also a key challenge. The time series information of remote sensing images can provide important clues for monitoring changes in land objects, but most existing deep learning models fail to effectively utilize this valuable resource. Finally, the lack of effective estimation of uncertainty also limits the application of the model. In practical applications, due to factors such as differences in data quality and imbalance in land object categories, the model prediction results may have high uncertainty, which in turn leads to poor accuracy and robustness in land object recognition.

[0005] Therefore, there is an urgent need for a more intelligent and adaptable method for extracting ground object information that can overcome the above shortcomings. Summary of the invention

[0006] In order to solve the above problems in the prior art, that is, to solve the problem that the prior art has poor accuracy and robustness when performing ground object information extraction and category recognition, the first aspect of the present invention proposes an intelligent ground object information extraction method based on remote sensing image big data, which is used to extract the features of ground objects and then perform category recognition. The method includes:

[0007] S1, obtaining remote sensing image data of ground objects to be classified as input data;

[0008] S2, preprocessing the input data to obtain preprocessed data;

[0009] S3, extracting features of the ground object based on the preprocessed data, and fusing the extracted features to obtain fused features;

[0010] S4, inputting the fusion feature into a trained ground object classification model to obtain a category recognition result of the ground object;

[0011] Among them, the land feature classification model is constructed based on a deep learning model.

[0012] In some preferred embodiments, the preprocessed data includes radiation correction, geometric correction, and atmospheric correction;

[0013] The method for performing radiation correction on the remote sensing image data is as follows:

[0014] Converting digital values ​​corresponding to remote sensing image data recorded by the sensor into physical quantities; the physical quantities include radiance;

[0015] Calculating a difference between the digital value and a minimum quantized value of the digital value as a first difference;

[0016] Calculate the difference between the maximum possible physical quantity and the minimum possible physical quantity as a second difference; calculate the difference between the maximum quantized value of the digital value and the minimum quantized value of the digital value as a third difference;

[0017] The second difference is divided by the third difference, the third difference is multiplied by the first difference after the division, and the third difference is added to the minimum possible physical quantity to obtain a radiation correction result.

[0018] In some preferred embodiments, if the physical quantity is radiance, atmospheric correction is performed on the remote sensing image data by:

[0019]

[0020] Where ρ represents the surface reflectivity, L represents the radiation correction result, and L prepresents the radiance of the atmospheric path, E0 represents the irradiance of the outer solar space, θ s represents the solar zenith angle, T s , T v They represent the atmospheric transmittance in the direction of the sun and the atmospheric transmittance in the observation direction respectively.

[0021] In some preferred embodiments, the geometric correction includes internal distortion correction and external geometric correction.

[0022] In some preferred embodiments, the features of the ground object are extracted, and the extracted features are fused to obtain fused features, and the method is as follows:

[0023] Respectively performing spectral feature extraction, texture feature extraction, shape feature extraction, and spatial relationship feature extraction on the ground objects;

[0024] The extracted spectral features, texture features, shape features, and spatial relationship features are weighted fused to obtain fused features.

[0025] In some preferred embodiments, the object classification model includes a spatial attention mechanism, a depth-separable convolution, a multi-scale feature pyramid, an autoencoder, and an output layer;

[0026] The importance weight of the fused feature at each spatial position is calculated by the spatial attention mechanism, and weighted, and the weighted feature is used as the first feature;

[0027] The deep classifiable convolution performs a deep convolution on each input channel of the first feature, and then uses the convolution-processed feature as the second feature through point-by-point convolution;

[0028] Inputting the second feature into the multi-scale feature pyramid for feature enhancement processing, and using the enhanced feature as the third feature;

[0029] Performing automatic encoding and decoding processing on the third feature by an automatic encoder to obtain a reconstructed feature;

[0030] The reconstructed features are input into the output layer, and full connection and softmax processing are performed in sequence to obtain the category recognition result.

[0031] In some preferred embodiments, the loss function of the ground object classification model during training is:

[0032] L MTL =-∑y i log(p i )+α(1-p i )log(p i )

[0033] Among them, L MTL Represents the loss value, y i represents the true value label, p i It represents the predicted value of the category recognition result, and α represents the weight.

[0034] In a second aspect of the present invention, an intelligent ground object information extraction system based on remote sensing image big data is proposed, which is used to extract the features of ground objects and then perform category recognition. The system includes:

[0035] A data acquisition module is configured to acquire remote sensing image data of ground objects to be classified as input data;

[0036] A data preprocessing module, configured to preprocess the input data to obtain preprocessed data;

[0037] A feature extraction module, configured to extract features of the ground object based on the preprocessed data, and fuse the extracted features to obtain a fused feature;

[0038] A ground object recognition module, configured to input the fusion feature into a trained ground object classification model to obtain a category recognition result of the ground object;

[0039] Among them, the land feature classification model is constructed based on a deep learning model.

[0040] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0041] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned intelligent ground feature information extraction method based on remote sensing image big data.

[0042] In a fourth aspect of the present invention, a computer-readable storage medium is proposed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by a computer to implement the above-mentioned intelligent ground object information extraction method based on remote sensing image big data.

[0043] Beneficial effects of the present invention:

[0044] 1) The present invention significantly improves the accuracy of ground feature information extraction and category recognition through multi-source data fusion and multi-scale feature extraction;

[0045] 2) Combining the attention mechanism and spatiotemporal feature enhancement, the model can better adapt to complex scenes and various types of objects, and improve the robustness of object information extraction and category recognition;

[0046] 3) The lightweight design reduces computing resource consumption, enabling the model to be quickly deployed and run in resource-limited environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings.

[0048] Figure 1 It is a flow chart of an intelligent ground feature information extraction method based on remote sensing image big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0051] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0052] The first embodiment of the present invention is an intelligent ground object information extraction method based on remote sensing image big data, which is used to extract the features of ground objects and then perform category recognition, such as Figure 1 As shown, the following steps are included:

[0053] S1, obtaining remote sensing image data of ground objects to be classified as input data;

[0054] S2, preprocessing the input data to obtain preprocessed data;

[0055] S3, extracting features of the ground object based on the preprocessed data, and fusing the extracted features to obtain fused features;

[0056] S4, inputting the fusion feature into a trained ground object classification model to obtain a category recognition result of the ground object;

[0057] Among them, the land feature classification model is constructed based on a deep learning model.

[0058] In order to more clearly illustrate the intelligent ground feature information extraction method based on remote sensing image big data of the present invention, each step in an embodiment of the method of the present invention is described in detail below with reference to the accompanying drawings.

[0059] S1, obtaining remote sensing image data of ground objects to be classified as input data;

[0060] In this embodiment, different types of data sources such as optical sensors (such as Landsat, Sentinel-2) and synthetic aperture radar (SAR) are obtained and integrated, and images covering the same area but at different times are collected as input images.

[0061] S2, preprocessing the input data to obtain preprocessed data;

[0062] In this embodiment, radiation correction, geometric correction, atmospheric correction, etc. are performed on the original remote sensing image data to eliminate various interference factors in the imaging process and ensure the image quality.

[0063] Radiation correction is used to eliminate the image caused by differences in sensor characteristics. The correction process is:

[0064] Converting digital values ​​corresponding to remote sensing image data recorded by the sensor into physical quantities; the physical quantities include radiance and emissivity;

[0065] Calculating a difference between the digital value and a minimum quantized value of the digital value as a first difference;

[0066] Calculate the difference between the maximum possible physical quantity and the minimum possible physical quantity as a second difference; calculate the difference between the maximum quantized value of the digital value and the minimum quantized value of the digital value as a third difference;

[0067] The second difference is divided by the third difference, the third difference is multiplied by the first difference after the division, and the third difference is added to the minimum possible physical quantity to obtain a radiation correction result.

[0068] Taking radiance as an example, the digital value recorded by the sensor is converted into radiance, and the correction formula is:

[0069]

[0070] Where L represents the radiation correction result, L max , L min Indicates the maximum possible radiance and the minimum possible radiance, Q col,max , Q col,min It represents the corresponding maximum quantization value and minimum quantization value (that is, the maximum quantization value and minimum quantization value that the sensor can record), and DN represents the digital value recorded by the sensor.

[0071] Atmospheric correction is used to remove the influence of atmospheric scattering and absorption on remote sensing image data and obtain the true surface response rate. The method for atmospheric correction of remote sensing image data is as follows:

[0072]

[0073] Where ρ represents the surface reflectivity, L represents the radiation correction result, and L p represents the radiance of the atmospheric path, E0 represents the irradiance of the outer solar space, θ s represents the solar zenith angle, T s , T v They represent the atmospheric transmittance in the direction of the sun and the atmospheric transmittance in the observation direction respectively.

[0074] Geometric correction is used to correct image deformation caused by factors such as sensor posture, terrain distortion, and earth curvature, so that the image pixel position accurately reflects the ground geographic coordinates. Geometric correction mainly includes internal distortion correction and external geometric correction.

[0075] Internal distortion correction is to correct internal factors such as lens distortion through sensor models or internal reference files; external geometric correction is to perform precise registration through ground control points. Both are existing technologies and will not be described here one by one.

[0076] S3, extracting features of the ground object based on the preprocessed data, and fusing the extracted features to obtain fused features;

[0077] In this embodiment, the feature extractor is used to extract spectral features, texture features, shape features, and spatial relationship features of ground objects respectively;

[0078] The extracted spectral features, texture features, shape features, and spatial relationship features are weightedly fused to obtain fused features and a rich feature representation.

[0079] S4, inputting the fusion features into a trained ground object classification model to obtain a category recognition result of the ground object; wherein the ground object classification model is constructed based on a deep learning model.

[0080] In this embodiment, the object classification model includes a spatial attention mechanism, a depth-separable convolution, a multi-scale feature pyramid, an autoencoder, and an output layer;

[0081] First, add a spatial attention mechanism to the front end of the model, dynamically adjust the weight according to the importance of each position, highlight the key area, and suppress noise interference. This helps to improve the accuracy of target detection in complex backgrounds. Specifically: the importance weight of the fused feature at each spatial position is calculated through the spatial attention mechanism, and an attention map with the same size as the input is generated and weighted. The weighted feature map has the same shape as the input, but highlights the important area, and the weighted feature is used as the first feature;

[0082] Then, depthwise separable convolution is used instead of standard convolution to reduce the number of parameters and speed up the calculation, thus maintaining the feature extraction capability while reducing the computational cost. This structure first applies convolution (depthwise convolution) to each input channel of the first feature independently, and then combines the results through pointwise convolution, and uses the convolution-processed features as the second feature.

[0083] After convolution processing, a multi-scale feature pyramid is constructed, the second feature is input, and the feature map is gradually fused from low to high levels, so that the final feature map contains multi-level information from local details to global structure, ensuring that the model can effectively capture information from local details to global structure. This is particularly useful for the classification of objects that require fine classification.

[0084] After enhancing the features through a multi-scale feature pyramid, an autoencoder is introduced to learn general feature representations using a large number of unlabeled remote sensing images, and then the classifier for a specific task is fine-tuned. This method can help the model gain better generalization capabilities in the case of small samples. Specifically: the third feature is first compressed through the encoder, and then restored through the decoder. The reconstructed features retain the main structure and pattern of the original features, enhancing the generalization ability;

[0085] The reconstructed features are input into the output layer, and the full connection and softmax processing are performed in sequence to obtain the category recognition result. That is, the full connection layer further integrates the features, and a dropout layer may be included to prevent overfitting. Finally, the softmax activation function is applied to generate the category probability distribution to achieve accurate classification.

[0086] The loss function of the object classification model during training is:

[0087] L MTL =-∑y i log(p i )+α(1-p i )log(p i )

[0088] Among them, L MTL Represents the loss value, yi represents the true value label, p i It represents the predicted value of the category recognition result, and α represents the weight.

[0089] In addition, in order to facilitate further understanding of the method of the present invention, the present invention takes agricultural monitoring as an example to illustrate the process of identifying the land object category. There are one hundred hectares of arable land in the North China Plain, where a variety of crops such as wheat, corn, soybeans, etc. are planted. Through the land object information extraction method of the present invention, multi-source remote sensing images taken by optical sensors (such as Sentinel-2), synthetic aperture radar (SAR) and unmanned aerial vehicles can be automatically processed, and the acquired multi-source remote sensing images are sequentially preprocessed, feature extracted, and feature fused, and then input into the trained land object classification model to efficiently and accurately identify and classify information such as different types of crops, thereby improving the accuracy of crop classification and helping to better plan field operations.

[0090] In summary, on the one hand, the present invention significantly improves the accuracy of object classification through multi-source data fusion and multi-scale feature extraction. On the other hand, it combines the attention mechanism and spatiotemporal feature enhancement to make the model more adaptable to complex scenes and various types of objects, thereby improving the robustness of object information extraction and recognition.

[0091] A second embodiment of the present invention is an intelligent ground object information extraction system based on remote sensing image big data, which is used to extract features of ground objects and then perform category recognition. The system includes:

[0092] A data acquisition module is configured to acquire remote sensing image data of ground objects to be classified as input data;

[0093] A data preprocessing module is configured to preprocess the input data to obtain preprocessed data; the preprocessed data includes radiation correction, geometric correction, and atmospheric correction;

[0094] A feature extraction module, configured to extract features of the ground object based on the preprocessed data, and fuse the extracted features to obtain a fused feature;

[0095] A ground object recognition module, configured to input the fusion feature into a trained ground object classification model to obtain a category recognition result of the ground object;

[0096] Among them, the land feature classification model is constructed based on a deep learning model.

[0097] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0098] It should be noted that the intelligent ground feature information extraction system based on remote sensing image big data provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

[0099] An electronic device according to the third embodiment of the present invention comprises: at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned intelligent ground feature information extraction method based on remote sensing image big data.

[0100] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned intelligent ground feature information extraction method based on remote sensing image big data.

[0101] Technicians in the technical field can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the electronic device and computer-readable storage medium described above can refer to the corresponding process in the aforementioned method example and will not be repeated here.

[0102] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0103] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.

[0104] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0105] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An intelligent ground object information extraction method based on remote sensing image big data, used to extract the features of ground objects and then perform category recognition, characterized in that: The method comprises the following steps: S1, obtaining remote sensing image data of ground objects to be classified as input data; S2, preprocessing the input data to obtain preprocessed data; S3, extracting features of the ground object based on the preprocessed data, and fusing the extracted features to obtain fused features; S4, inputting the fusion feature into a trained ground object classification model to obtain a category recognition result of the ground object; Among them, the land feature classification model is constructed based on a deep learning model.

2. The intelligent ground feature information extraction method based on remote sensing image big data according to claim 1 is characterized in that: The preprocessed data includes radiation correction, geometric correction, and atmospheric correction; The method for performing radiation correction on the remote sensing image data is as follows: Converting digital values ​​corresponding to remote sensing image data recorded by the sensor into physical quantities; the physical quantities include radiance; Calculating a difference between the digital value and a minimum quantized value of the digital value as a first difference; Calculate the difference between the maximum possible physical quantity and the minimum possible physical quantity as a second difference; calculate the difference between the maximum quantized value of the digital value and the minimum quantized value of the digital value as a third difference; The second difference is divided by the third difference, the third difference is multiplied by the first difference after the division, and the third difference is added to the minimum possible physical quantity to obtain a radiation correction result.

3. The intelligent ground feature information extraction method based on remote sensing image big data according to claim 2 is characterized in that: If the physical quantity is radiance, atmospheric correction is performed on the remote sensing image data in the following method: Where ρ represents the surface reflectivity, L represents the radiation correction result, and L p represents the radiance of the atmospheric path, E0 represents the irradiance of the outer solar space, θ s represents the solar zenith angle, T s , T v They represent the atmospheric transmittance in the direction of the sun and the atmospheric transmittance in the observation direction respectively.

4. The intelligent ground feature information extraction method based on remote sensing image big data according to claim 2 is characterized in that: The geometric correction includes internal distortion correction and external geometric correction.

5. The intelligent ground feature information extraction method based on remote sensing image big data according to claim 1 is characterized in that: The features of the ground object are extracted, and the extracted features are fused to obtain fused features, and the method is as follows: Respectively performing spectral feature extraction, texture feature extraction, shape feature extraction, and spatial relationship feature extraction on the ground objects; The extracted spectral features, texture features, shape features, and spatial relationship features are weighted fused to obtain fused features.

6. The intelligent ground feature information extraction method based on remote sensing image big data according to claim 5 is characterized in that: The object classification model includes a spatial attention mechanism, a deep separable convolution, a multi-scale feature pyramid, an autoencoder, and an output layer; The importance weight of the fused feature at each spatial position is calculated by the spatial attention mechanism, and weighted, and the weighted feature is used as the first feature; The deep classifiable convolution performs a deep convolution on each input channel of the first feature, and then uses the convolution-processed feature as the second feature through point-by-point convolution; Inputting the second feature into the multi-scale feature pyramid for feature enhancement processing, and using the enhanced feature as the third feature; Performing automatic encoding and decoding processing on the third feature by an automatic encoder to obtain a reconstructed feature; The reconstructed features are input into the output layer, and full connection and softmax processing are performed in sequence to obtain the category recognition result.

7. The intelligent ground feature information extraction method based on remote sensing image big data according to claim 6 is characterized in that: The loss function of the object classification model during training is: L MTL =-∑y i log(p i )+α(1-p i )log(p i ) Among them, L MTL Represents the loss value, y i represents the true value label, p i It represents the predicted value of the category recognition result, and α represents the weight.

8. An intelligent ground object information extraction system based on remote sensing image big data, used to extract the features of ground objects and then perform category recognition, characterized in that: The system includes: A data acquisition module is configured to acquire remote sensing image data of ground objects to be classified as input data; A data preprocessing module, configured to preprocess the input data to obtain preprocessed data; A feature extraction module, configured to extract features of the ground object based on the preprocessed data, and fuse the extracted features to obtain a fused feature; A ground object recognition module, configured to input the fusion feature into a trained ground object classification model to obtain a category recognition result of the ground object; Among them, the land feature classification model is constructed based on a deep learning model.

9. An electronic device, characterized in that: include: at least one processor; And a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement an intelligent ground feature information extraction method based on remote sensing image big data as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by a computer to implement the intelligent ground feature information extraction method based on remote sensing image big data as described in any one of claims 1-7.

Images