Ship target multi-dimensional feature extraction method based on satellite-borne SAR image

Abstract

The invention discloses a ship target multi-dimensional feature extraction method based on a satellite-borne SAR image, and the method is used for a satellite ground station data processing center and / or a satellite data processing center. The data processing center comprises an image data read-in module and a plurality of image feature extraction modules, and the method comprises the steps that the image data read-in module receives a satellite-borne SAR ship target slice image and generates a to-be-processed task image; extracting image spatial domain features and image time domain features from the task image to be processed; image frequency domain features are obtained; and the output module respectively receives the feature data output by the image feature extraction module, analyzes and merges the feature data, and generates feature descriptions of multi-dimensional information of three description types of texts, images and numerical values.

Description

Technical Field

[0001] The present invention relates to the technical field of radar image processing, and in particular to a method for extracting multi-dimensional features of ship targets based on satellite-borne SAR images. Background Art

[0002] Synthetic Aperture Radar (SAR) is an imaging radar that uses the relative motion between the radar and the target to synthesize a smaller real antenna aperture into a larger equivalent antenna aperture through data processing. Synthetic aperture radar can achieve high resolution in both range and azimuth, and has the advantages of all-day, all-weather, camouflage recognition and penetration of cover, and is widely used in various fields. In agriculture, it is used in many fields such as crop identification and classification, agricultural soil moisture inversion, and crop growth monitoring. In forestry, SAR can complete tasks such as forest source surveys, forestry zoning and forest classification, and forest natural disaster detection. More importantly, in the military, it is mainly used to detect and identify specific military targets.

[0003] As a maritime power, my country has a long coastline, vast territorial waters and exclusive economic zones. Using satellite-borne SAR to detect and identify sea targets is of great significance to safeguarding maritime rights and interests and ensuring national security. In the military field, it can obtain information such as the location, direction, and speed of sea targets without being affected by weather and time periods, thereby supporting military operations. In the civilian field, it is conducive to better maritime rescue, strengthening water transportation management in specific bays, ports and sea areas, and combating illegal fishing.

[0004] Chinese patent publication CN107895139A discloses a SAR image target recognition method based on multi-feature fusion, the method comprising: extracting edges from SAR images, and then obtaining the compactness, fullness and complexity of the target; determining the gray-level co-occurrence matrix of the SAR image, and using the gray-level co-occurrence matrix to calculate texture feature quantities; constructing a multi-layer superpixel set for the SAR image, and constructing an unbalanced bidirectional graph based on the multi-layer superpixel set, and separating the image target from the background by clustering; calculating the covariance matrix using the fused image feature matrix, and constructing an optimal projection matrix, projecting the training samples onto the optimal projection matrix, and obtaining samples after dimensionality reduction; training the final target identifier, and in each round of training, using the weights of the eigenvalues ​​of different training data samples to statistically weakly classify, and selecting weak classifiers according to the different classification error rates of each eigenvalue, and weighted summing the weak classifiers to construct an output classifier. However, the patent disclosed in this patent is still a technical solution for one-dimensional feature recognition.

[0005] The extraction of maritime target image features is the key to realizing the recognition of maritime target ships by spaceborne SAR. Target features are the key to determining similarity and classification. The essence is a dimensionality reduction of the image. Using a specific method, certain characteristics of the image can be extracted to reflect the essence of the image. Ideally, the features are complementary and complete. The dimension of the feature is often lower than the original dimension of the image, so it is more suitable for image classification and recognition, which can speed up the calculation speed and improve efficiency. Multi-dimensional target features can significantly improve the accuracy and timeliness of back-end interpretation tasks such as target detection and recognition. How to process maritime target images to generate multi-dimensional features is an urgent problem to be solved. Summary of the invention

[0006] A first aspect of the present invention provides a method for extracting multi-dimensional features of ship targets based on spaceborne synthetic aperture radar (SAR) images, which is applied to a satellite ground station data processing center and / or a satellite onboard data processing center. The data processing center comprises: an image data reading module, a first image feature extraction module, a second image feature extraction module, a third image feature extraction module and an output module. The method comprises the following steps:

[0007] Step 1, the image data reading module receives the ship image of the spaceborne SAR, divides the image into a target slice image sequence according to a specified resolution, and generates a task image to be processed from the target slice image sequence;

[0008] Step 2, the first image feature extraction module receives the task image to be processed, and extracts image spatial domain features from the task image to be processed;

[0009] Step 3, the second image feature extraction module receives the task image to be processed, and extracts image time domain features from the task image to be processed;

[0010] Step 4: The third image feature extraction module receives the task image to be processed and extracts image frequency domain features from the task image to be processed;

[0011] Step 5, the output module receives the feature data output by the first image feature extraction module, the second image feature extraction module and the third image feature extraction module respectively, performs analysis and merging processing on the feature data, and generates feature descriptions of multidimensional information of three description types: text, image and numerical value.

[0012] As described in the method of the first aspect of the present invention, step 1 further comprises the sub-steps of:

[0013] Step 1.1, the image data reading module receives the ship image of the spaceborne SAR and stores it in the image storage unit area of ​​the data processing center memory in chronological order;

[0014] Step 1.2, the image data reading module extracts the stored intelligent detection algorithm from the data processing center memory, and uses the intelligent detection algorithm to slice the ship image according to the specified resolution to generate a target slice image;

[0015] Step 1.3, the image data reading module stores the target slice image in the pending task storage unit area of ​​the data processing center memory again in chronological order;

[0016] Step 1.4, the image data reading module extracts and outputs the target slice image from the pending task storage unit area as the pending task.

[0017] As described in the method of the first aspect of the present invention, step 1.2 also includes: for images that cannot be processed by the intelligent detection algorithm or images that cannot be recognized after processing, downloading them to an image processing device to perform manual interpretation and slicing to generate target slice images, and storing the target slice images in a data processing center memory.

[0018] As described in the method of the first aspect of the present invention, the specified resolution in step 1.2 includes: dividing the ship image according to a predetermined number of pixels of length × width, and the predetermined number of pixels in the length direction and the width direction is 800-1000 pixels; adjacent image slices are set with overlapping areas according to the predetermined number of pixels of the image, and the distance of the overlapping areas is 100-300 pixels.

[0019] According to the method of the first aspect of the present invention, step 2 comprises the following sub-steps:

[0020] Step 2.1: The first image feature extraction module receives the target slice image to be processed and obtains the spatial feature extraction processing algorithm from the data processing center memory;

[0021] Step 2.2, the first image feature extraction module processes the task to be processed using a spatial feature extraction algorithm based on a Canny operator, a SIFT operator and a LBP operator to extract image spatial features;

[0022] The image spatial domain features include: edge features of the target image in the spatial domain, scale-invariant features of the target image, and texture features of the target image.

[0023] According to the method of the first aspect of the present invention, step 3 comprises the following sub-steps:

[0024] Step 3.1, the second image feature extraction module receives the task to be processed of the target slice image, and obtains the time domain feature extraction processing algorithm from the data processing center memory;

[0025] Step 3.2, the second image feature extraction module uses the time domain feature extraction processing algorithm to process the task to be processed to generate a frequency spectrum of the target slice image in the azimuth direction;

[0026] Step 3.3, the second image feature extraction module divides the spectrum in the azimuth direction according to the sub-apertures, and performs imaging on the divided spectrum in the azimuth direction respectively to obtain imaging results of each sub-aperture;

[0027] Step 3.4: The second image feature extraction module generates an image sequence according to the time sequence of each sub-aperture imaging result, and generates a feature description of the target slice image in the time domain dimension.

[0028] As described in the method of the first aspect of the present invention, step 3.2 comprises: using an azimuthal inverse Fourier transform on the task to be processed to obtain a frequency spectrum of the target slice image in the azimuthal direction.

[0029] According to the method of the first aspect of the present invention, step 4 comprises the following sub-steps:

[0030] Step 4.1, the third image feature extraction module receives the task to be processed of the target slice image, and obtains the frequency domain feature extraction processing algorithm from the data processing center memory;

[0031] Step 4.2, the third image feature extraction module uses Fourier transform on the task to be processed to obtain high-frequency and low-frequency features of the target slice image, and generates a frequency domain map of the task to be processed;

[0032] Step 4.3, the third image feature extraction module processes the task to be processed using a wavelet transform operator to generate a wavelet domain decomposition graph

[0033] Step 4.4: The third image feature extraction module extracts a feature description of the target slice image in the frequency domain dimension from the frequency domain image and the wavelet domain decomposition image of the task to be processed.

[0034] As described in the first aspect of the present invention, step 5 includes: analyzing and merging the spatial, temporal, and frequency domain feature items of the target slice images extracted in steps 2, 3, and 4, generating a multi-dimensional feature set of ship targets of SAR images including three description types of text, image, and numerical value, covering the three feature dimensions of spatial, temporal, and frequency domains.

[0035] The second aspect of the present invention provides a multi-dimensional feature extraction device for ship targets based on space-borne synthetic aperture radar SAR images, comprising: an image data reading module, a first image feature extraction module, a second image feature extraction module, a third image feature extraction module and an output module, characterized in that the image data reading module receives the ship image of the space-borne SAR, divides the image into a target slice image sequence according to a specified resolution, and generates a task image to be processed; the first image feature extraction module receives the task image to be processed, and extracts image spatial domain features from the task image to be processed; the second image feature extraction module receives the task image to be processed, and extracts image time domain features from the task image to be processed; the third image feature extraction module receives the task image to be processed, and extracts image frequency domain features from the task image to be processed; the output module respectively receives the feature data output by the first image feature extraction module, the second image feature extraction module and the third image feature extraction module, performs analysis and merging processing on the feature data, and generates a multi-dimensional feature set of three description types: text, image and numerical value.

[0036] The method of the present invention has the following advantages:

[0037] (1) The present invention proposes a multi-dimensional feature extraction method for ship targets based on spaceborne SAR images, which involves feature acquisition methods at multiple scales, such as spaceborne SAR imaging principle, multi-domain transformation, and image processing. Compared with the traditional single-dimensional feature extraction method, the method has a higher coverage dimension, obtains more comprehensive feature information, and generates more standardized feature descriptions, further improving the ability to describe target features.

[0038] (2) The multi-dimensional feature extraction method for ship targets based on satellite-borne SAR images proposed in the present invention has been tested on multiple SAR open source datasets such as SSDD, SSDD+, and MTCD, and has been verified on multiple actual model SAR satellite ship target data such as GF-3 and Haisi-1. The method has good universality, improves the target feature description capability, and improves the standardization of target feature description.

[0039] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following is a preferred embodiment, which is described in detail with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Flowchart of the method for extracting multi-dimensional features of ship targets based on spaceborne SAR images proposed in the present invention. DETAILED DESCRIPTION

[0041] The present invention discloses a multi-dimensional feature extraction method for ship targets based on spaceborne synthetic aperture radar (SAR) images. Aiming at the task of extracting and analyzing features of marine ship targets from spaceborne SAR images, the present invention aims at the problem that a single feature extraction method obtains low information dimension and insufficient information quantity, and proposes a multi-dimensional feature extraction method for ship targets based on spaceborne SAR images based on multiple types of remote sensing SAR image feature extraction and processing algorithms.

[0042] The technical scheme of the present invention will be clearly and completely described below in conjunction with specific embodiments, but those skilled in the art should understand that the embodiments described below are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention. 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.

[0043] A first aspect of the present invention provides a method for extracting multi-dimensional features of ship targets based on spaceborne synthetic aperture radar (SAR) images, which is applied to a satellite ground station data processing center and / or a satellite onboard data processing center. The data processing center comprises: an image data reading module, a first image feature extraction module, a second image feature extraction module, a third image feature extraction module and an output module. The method comprises the following steps:

[0044] Step 1, the image data reading module receives the ship image of the spaceborne SAR, divides the image into a target slice image sequence according to a specified resolution, and generates a task image to be processed from the target slice image sequence;

[0045] Step 2, the first image feature extraction module receives the task image to be processed, and extracts image spatial domain features from the task image to be processed;

[0046] Step 3, the second image feature extraction module receives the task image to be processed, and extracts image time domain features from the task image to be processed;

[0047] Step 4: The third image feature extraction module receives the task image to be processed and extracts image frequency domain features from the task image to be processed;

[0048] Step 5, the output module receives the feature data output by the first image feature extraction module, the second image feature extraction module and the third image feature extraction module respectively, performs analysis and merging processing on the feature data, and generates feature descriptions of multidimensional information of three description types: text, image and numerical value.

[0049] As described in the method of the first aspect of the present invention, step 1 further comprises the sub-steps of:

[0050] Step 1.1, the image data reading module receives the ship image of the spaceborne SAR and stores it in the image storage unit area of ​​the data processing center memory in chronological order;

[0051] Step 1.2, the image data reading module extracts the stored intelligent detection algorithm from the data processing center memory, and uses the intelligent detection algorithm to slice the ship image according to the specified resolution to generate a target slice image;

[0052] Step 1.3, the image data reading module stores the target slice image in the pending task storage unit area of ​​the data processing center memory again in chronological order;

[0053] Step 1.4, the image data reading module extracts and outputs the target slice image from the pending task storage unit area as the pending task.

[0054] Example 1

[0055] The technical problem to be solved by the present invention is to obtain a ship image as an identification target from a received satellite synthetic aperture radar (SAR) sea surface image, and extract the spatial domain, time domain and frequency domain features of the ship for image recognition. The image format of the satellite synthetic aperture radar SAR obtained from the beamforming or stripe mode of the GF-3 is very large, usually a stripe-shaped image, and first of all, the image with the ship target must be found from the image of the synthetic aperture radar SAR. The method of the present invention is to firstly divide the large-format high-pixel image into many small-format images of specified pixels by an image segmentation method, such as an image with a size of 800×800 pixels. Of course, the number of pixels of this rectangular image can be larger or smaller, such as 800-1000 pixels in the length or width direction of the image, and there can be an overlapping area between each adjacent image, and the image resolution sets the overlapping area to 100 to 300 pixels, typically 100 pixels. Then, the sea surface image without the ship target to be identified is excluded to generate the target slice image to be detected. To exclude images without target ships, a method based on grayscale histogram statistics can be performed for each target slice, and images with constant grayscale can be excluded.

[0056] As described in the method of the first aspect of the present invention, step 1.2 also includes: for images that cannot be processed by the intelligent detection algorithm or images that cannot be recognized after processing, downloading them to an image processing device to perform manual interpretation and slicing to generate target slice images, and storing the target slice images in a data processing center memory.

[0057] If the target slice image cannot be accurately obtained by using the above intelligent detection algorithm, the image can also be downloaded to an image display device for human interpretation and graphic slicing operations to obtain the target slice image for the task to be processed.

[0058] As described in the method of the first aspect of the present invention, the specified resolution in step 1.2 includes: dividing the ship image according to a predetermined number of pixels of length × width, and the predetermined number of pixels in the length direction and the width direction is 800-1000 pixels; adjacent image slices are set with overlapping areas according to the predetermined number of pixels of the image, and the distance of the overlapping areas is 100-300 pixels.

[0059] According to the method of the first aspect of the present invention, step 2 comprises the following sub-steps:

[0060] Step 2.1: The first image feature extraction module receives the target slice image to be processed and obtains the spatial feature extraction processing algorithm from the data processing center memory;

[0061] Step 2.2, the first image feature extraction module processes the task to be processed using a spatial feature extraction algorithm based on a Canny operator, a SIFT operator and a LBP operator to extract image spatial features;

[0062] The image spatial domain features include: edge features of the target image in the spatial domain, scale-invariant features of the target image, and texture features of the target image.

[0063] The spatial feature extraction algorithm can be used to first use an algorithm based on the Canny operator to process the task to be processed, then use an algorithm based on the SIFT operator to process it, and finally use an algorithm based on the LBP operator to process it.

[0064] According to the method of the first aspect of the present invention, step 3 comprises the following sub-steps:

[0065] Step 3.1, the second image feature extraction module receives the task to be processed of the target slice image, and obtains the time domain feature extraction processing algorithm from the data processing center memory;

[0066] Step 3.2, the second image feature extraction module uses the time domain feature extraction processing algorithm to process the task to be processed to generate a frequency spectrum of the target slice image in the azimuth direction;

[0067] Step 3.3, the second image feature extraction module divides the spectrum in the azimuth direction according to the sub-apertures, and performs imaging on the divided spectrum in the azimuth direction respectively to obtain imaging results of each sub-aperture;

[0068] Step 3.4: The second image feature extraction module generates an image sequence according to the time sequence of each sub-aperture imaging result, and generates a feature description of the target slice image in the time domain dimension.

[0069] As described in the method of the first aspect of the present invention, step 3.2 comprises: using an azimuthal inverse Fourier transform on the task to be processed to obtain a frequency spectrum of the target slice image in the azimuthal direction.

[0070] The time domain feature extraction and processing algorithm first performs an inverse Fourier transform on the target slice image in the azimuth direction to obtain the spectrum of the target slice image in the azimuth direction; then performs azimuth segmentation on the image, and finally performs sub-aperture imaging.

[0071] According to the method of the first aspect of the present invention, step 4 comprises the following sub-steps:

[0072] Step 4.1, the third image feature extraction module receives the task to be processed of the target slice image, and obtains the frequency domain feature extraction processing algorithm from the data processing center memory;

[0073] Step 4.2, the third image feature extraction module uses Fourier transform on the task to be processed to obtain high-frequency and low-frequency features of the target slice image, and generates a frequency domain map of the task to be processed;

[0074] Step 4.3, the third image feature extraction module processes the task to be processed using a wavelet transform operator to generate a wavelet domain decomposition graph

[0075] Step 4.4: The third image feature extraction module extracts a feature description of the target slice image in the frequency domain dimension from the frequency domain image and the wavelet domain decomposition image of the task to be processed.

[0076] As described in the first aspect of the present invention, step 5 includes: analyzing and merging the spatial, temporal, and frequency domain feature items of the target slice images extracted in steps 2, 3, and 4, generating a multi-dimensional feature set of ship targets of SAR images including three description types of text, image, and numerical value, covering the three feature dimensions of spatial, temporal, and frequency domains.

[0077] The second aspect of the present invention provides a multi-dimensional feature extraction device for ship targets based on space-borne synthetic aperture radar SAR images, comprising: an image data reading module, a first image feature extraction module, a second image feature extraction module, a third image feature extraction module and an output module, characterized in that the image data reading module receives the ship image of the space-borne SAR, divides the image into a target slice image sequence according to a specified resolution, and generates a task image to be processed; the first image feature extraction module receives the task image to be processed, and extracts image spatial domain features from the task image to be processed; the second image feature extraction module receives the task image to be processed, and extracts image time domain features from the task image to be processed; the third image feature extraction module receives the task image to be processed, and extracts image frequency domain features from the task image to be processed; the output module respectively receives the feature data output by the first image feature extraction module, the second image feature extraction module and the third image feature extraction module, performs analysis and merging processing on the feature data, and generates a multi-dimensional feature set of three description types: text, image and numerical value.

[0078] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for extracting multi-dimensional features of ship targets based on spaceborne synthetic aperture radar (SAR) images, applied to a satellite ground station data processing center and / or a satellite onboard data processing center, the data processing center comprising: The image data reading module, the first image feature extraction module, the second image feature extraction module, the third image feature extraction module and the output module are characterized in that the method comprises the following steps: Step 1, the image data reading module receives the ship image of the spaceborne SAR, divides the image into a target slice image sequence according to a specified resolution, and generates a task image to be processed from the target slice image sequence; Step 2, the first image feature extraction module receives the task image to be processed, and extracts image spatial domain features from the task image to be processed; Step 3, the second image feature extraction module receives the task image to be processed, and extracts image time domain features from the task image to be processed; Step 4: The third image feature extraction module receives the task image to be processed and extracts image frequency domain features from the task image to be processed; Step 5, the output module receives the feature data output by the first image feature extraction module, the second image feature extraction module and the third image feature extraction module respectively, performs analysis and merging processing on the feature data, and generates feature descriptions of multidimensional information of three description types: text, image and numerical value.

2. The method according to claim 1, characterized in that Step 1 also includes the following sub-steps: Step 1.1, the image data reading module receives the ship image of the spaceborne SAR and stores it in the image storage unit area of ​​the data processing center memory in chronological order; Step 1.2, the image data reading module extracts the stored intelligent detection algorithm from the data processing center memory, and uses the intelligent detection algorithm to slice the ship image according to the specified resolution to generate a target slice image; Step 1.3, the image data reading module stores the target slice image in the pending task storage unit area of ​​the data processing center memory again in chronological order; Step 1.4, the image data reading module extracts and outputs the target slice image from the pending task storage unit area as the pending task.

3. The method according to claim 2, characterized in that The step 1.2 also includes: for images that cannot be processed by the intelligent detection algorithm or images that cannot be recognized after processing, downloading them to an image processing device for manual interpretation and slicing to generate target slice images, and storing the target slice images in a data processing center memory.

4. The method according to claim 2, characterized in that The specified resolution in step 1.2 includes: dividing the ship image according to a predetermined number of pixels of length × width, and the predetermined number of pixels in the length direction and the width direction is 800-1000 pixels; adjacent image slices are set with overlapping areas according to the predetermined number of pixels of the image, and the distance of the overlapping areas is 100-300 pixels.

5. The method according to claim 1, characterized in that Step 2 includes the following sub-steps: Step 2.1: The first image feature extraction module receives the target slice image to be processed and obtains the spatial feature extraction processing algorithm from the data processing center memory; Step 2.2, the first image feature extraction module processes the task to be processed using a spatial feature extraction algorithm based on a Canny operator, a SIFT operator and a LBP operator to extract image spatial features; The image spatial domain features include: edge features of the target image in the spatial domain, scale-invariant features of the target image, and texture features of the target image.

6. The method according to claim 1, characterized in that Step 3 includes the following sub-steps: Step 3.1, the second image feature extraction module receives the task to be processed of the target slice image, and obtains the time domain feature extraction processing algorithm from the data processing center memory; Step 3.2, the second image feature extraction module uses the time domain feature extraction processing algorithm to process the task to be processed to generate a frequency spectrum of the target slice image in the azimuth direction; Step 3.3, the second image feature extraction module divides the spectrum in the azimuth direction according to the sub-apertures, and performs imaging on the divided spectrum in the azimuth direction respectively to obtain imaging results of each sub-aperture; Step 3.4: The second image feature extraction module generates an image sequence according to the time sequence of each sub-aperture imaging result, and generates a feature description of the target slice image in the time domain dimension.

7. The method according to claim 6, characterized in that Step 3.2 includes: using an azimuthal inverse Fourier transform on the task to be processed to obtain a frequency spectrum of the target slice image in the azimuthal direction.

8. The method according to claim 1, characterized in that The step 4 includes the following sub-steps: Step 4.1, the third image feature extraction module receives the task to be processed of the target slice image, and obtains the frequency domain feature extraction processing algorithm from the data processing center memory; Step 4.2, the third image feature extraction module uses Fourier transform on the task to be processed to obtain high-frequency and low-frequency features of the target slice image, and generates a frequency domain map of the task to be processed; Step 4.3, the third image feature extraction module processes the task to be processed using a wavelet transform operator to generate a wavelet domain decomposition graph Step 4.4: The third image feature extraction module extracts a feature description of the target slice image in the frequency domain dimension from the frequency domain image and the wavelet domain decomposition image of the task to be processed.

9. The method according to claim 1, characterized in that Step 5 includes: analyzing and merging the spatial, temporal and frequency domain feature items of the target slice images extracted in steps 2, 3 and 4, generating a multi-dimensional feature set of ship targets of SAR images including three description types of text, image and numerical value, covering the three feature dimensions of spatial, temporal and frequency domains.

10. A multi-dimensional feature extraction device for ship targets based on spaceborne synthetic aperture radar SAR images, comprising: The invention relates to an image data reading module, a first image feature extraction module, a second image feature extraction module, a third image feature extraction module and an output module, wherein the image data reading module receives a ship image of a spaceborne SAR, divides the image into a target slice image sequence according to a specified resolution, and generates a task image to be processed; the first image feature extraction module receives the task image to be processed, and extracts image spatial domain features from the task image to be processed; the second image feature extraction module receives the task image to be processed, and extracts image time domain features from the task image to be processed; the third image feature extraction module receives the task image to be processed, and extracts image frequency domain features from the task image to be processed; and the output module receives feature data output by the first image feature extraction module, the second image feature extraction module and the third image feature extraction module respectively, performs analysis and merging processing on the feature data, and generates a multidimensional feature set of three description types: text, image and numerical value.

Images