Multi-spectral image local feature extraction method and system based on deep learning

Abstract

The invention provides a multi-spectral image local feature extraction method and system based on deep learning, and belongs to the technical field of multi-spectral image processing, and the method comprises the steps: obtaining a pair of to-be-processed multi-spectral images; utilizing a spectrum selection network to extract shallow layer features from the pair of acquired multispectral images; processing the shallow layer features by using a feature extraction network, and generating a feature vector of each pixel point in the image; wherein in the training stage, a proposed loss function is used for training, and in the reasoning stage, descriptors are extracted according to matching requirements; wherein the nearest neighbor algorithm is adopted to carry out rough matching on the existing descriptors, and the RANSAC method is adopted to carry out filtering, so that a final matching result is obtained. According to the method, a spectrum selection network of a neural network channel is adaptively activated according to features, a plurality of images from different spectrums can be accepted at the same time for training, and meanwhile, a contrast loss function containing more negative samples is designed to extract multispectral image features.

Description

Technical Field

[0001] The present invention relates to the technical field of multispectral image processing, and in particular to a method and system for extracting local features of multispectral images based on deep learning. Background Art

[0002] Feature point matching is an important research problem in the field of computer vision and image processing, and is widely used in image retrieval, 3D reconstruction, positioning, and SLAM tasks. Manual feature matching is mainly divided into three steps: feature point extraction, calculation of main direction, and extraction of descriptors. Using the neighborhood area of ​​feature points to establish descriptors associated with feature points is the key issue of feature point matching. Based on artificially designed features, gradients and their variants are generally used, such as SIFT, SURF, ORB, etc. However, for multispectral images, gradient-based methods do not work well for the features of nonlinear radiation distortion (NRD) in multispectral images. In recent years, with the development of deep learning technology, local feature extraction technology based on deep learning has gradually been widely used.

[0003] However, existing algorithms perform poorly in extracting descriptors for multispectral images. Existing methods widely use pseudo-twin networks for multispectral feature extraction, which requires special training for different types of multispectral images, and the input image requires known image spectral classification.

[0004] Existing multispectral image feature extraction methods, based on manual feature descriptors such as SIFT, have insufficient abstraction capabilities and rely on features such as corners and gradients. However, due to the thermal diffusion effect of thermal infrared imaging, gradient smoothing and edge blurring may cause feature extraction failure in thermal infrared imaging. Existing deep learning-based feature extraction methods, such as CMMNet (Cross-Modality Matching Network) [4], use pseudo-twin networks to extract different spectral images separately when extracting feature images. For example, the basic image features of the light-far infrared image pair can be extracted. This method requires specific training for each multispectral image pair. In terms of feature extraction network training, the use of a simple ternary loss function for training is not effective.

[0005] The main technical problems of existing multispectral image feature extraction algorithms: Limitations of descriptors based on manual features: Traditional manual feature descriptors such as SIFT mainly rely on low-level features such as corners and gradients, and these methods have insufficient abstraction capabilities. Especially in thermal infrared imaging, these manual feature extraction methods may fail due to gradient smoothing and edge blurring caused by thermal diffusion effects. Specificity of feature extraction methods based on deep learning: Deep learning methods such as CMMNet (Cross-Modality Matching Network) use pseudo-twin networks to process images of different spectra (such as visible light and far-infrared image pairs) when extracting features. This method requires specific training for each multispectral image pair, which limits its generalization ability and application scope. Efficiency issues of feature extraction network training: When training the feature extraction network, only the ternary loss function containing anchor samples, positive samples and negative samples is used as training. When training multispectral features, it may cause the anchor samples to be closer to the positive samples and the anchor samples to be farther away from the negative samples, but the distance between the positive samples and the negative samples is not farther, or even closer, which will lead to registration failure. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for extracting local features of multispectral images based on deep learning, design a spectral selection network that adaptively activates neural network channels according to features, can simultaneously accept multiple images from different spectra for training, and design a contrast loss function containing more negative samples to extract multispectral image features, so as to solve at least one technical problem existing in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a method for extracting local features of a multispectral image based on deep learning, comprising:

[0009] Obtaining a pair of multispectral images to be processed;

[0010] The spectral selection network is used to extract shallow features from the acquired one-to-multispectral images;

[0011] The shallow features are processed by the feature extraction network to generate the feature vector of each pixel in the image; the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching results.

[0012] Furthermore, the image is input into the original backbone network to generate an adaptive gate function determined by the input image, and then the gate function is correspondingly multiplied to the filter of the convolutional network, activating only part of the convolution kernel of the network to extract the shallow features of the image.

[0013] Furthermore, the shallow features of the extracted image are:

[0014]

[0015] in, represents the feature map output of the ith channel of the ith layer, represents the i-th convolution kernel of the ,th layer, Indicates whether the i-th convolution kernel of the network layer is activated, and φ represents the nonlinear activation function; if not activated, the output is not 0, and if activated, the feature map I l (x) Perform convolution operation.

[0016] Furthermore, a loss function for training multi-spectral local features is constructed by imitating the information entropy loss function. When the image is input into the neural network, a vector with a length of 128 dimensions is generated for each point in the original image position.

[0017] Furthermore, during the training process, a spectral selection network is used to generate a descriptor with a length of 128 for each pixel in the multispectral image; after introducing more negative samples, the loss function can reduce the problems caused by the traditional triplet loss function and improve the accuracy of multispectral image registration.

[0018] Furthermore, the loss function L is:

[0019]

[0020] Where N is the total number of samples, z i represents the feature representation of the i-th sample, z i+ Indicates that z i Matching feature z i-j Represents negative samples between different graphs, z i-m Indicates that z i Adjacent negative samples in the same image, z i-n Indicates that z i+ For adjacent negative samples in the same image, τ, τ2, and τ3 represent the temperature values ​​of negative samples from different images and negative samples from the same image, respectively. The temperature values ​​are set to make the samples have better discrimination after training.

[0021] In a second aspect, the present invention provides a multispectral image local feature extraction system based on deep learning, comprising:

[0022] An acquisition module, used for acquiring a pair of multispectral images to be processed;

[0023] An extraction module, used for extracting shallow features from the acquired one-to-multi-spectral image using a spectral selection network;

[0024] The processing module is used to process the shallow features using the feature extraction network to generate the feature vector of each pixel in the image; wherein, the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; wherein, the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching result.

[0025] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the multispectral image local feature extraction method based on deep learning as described in the first aspect is implemented.

[0026] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the multispectral image local feature extraction method based on deep learning as described in the first aspect.

[0027] In a fifth aspect, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the multispectral image local feature extraction method based on deep learning as described in the first aspect.

[0028] Terminology explanation:

[0029] InfoNCE loss function: a common loss function in the field of contrastive learning.

[0030] Ternary loss function: It learns to distinguish the distance between an anchor sample and the positive sample and the distance between the anchor sample and the negative sample. It promotes the model to learn better feature representations that distinguish different categories by minimizing the distance between the anchor and the positive sample and maximizing the distance between the anchor and the negative sample.

[0031] RANSAC (RANdom Sample Consensus): Random sampling consensus method, filtering for matching points.

[0032] The beneficial effects of the present invention are as follows: a spectral selection network is designed to adaptively activate neural network channels according to features, which can simultaneously accept multiple images from different spectra for training, and a contrast loss function containing more negative samples is designed to extract multispectral image features.

[0033] Additional advantages of the present invention will be more clearly given in the following description or learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0035] Figure 1 This is a diagram of the multi-spectrum adaptive network structure described in an embodiment of the present invention.

[0036] Figure 2 A schematic diagram of the descriptor matching effect described in an embodiment of the present invention.

[0037] Figure 3 It is a schematic diagram of the structure of the spectral selection network (SpectralGate Net) described in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below by the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0039] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0040] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with that in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0041] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0042] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. Different embodiments or examples described in this specification and features of different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

[0043] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0044] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0045] Example 1

[0046] In this embodiment 1, a multispectral image local feature extraction system based on deep learning is first provided, including: an acquisition module, used to acquire a pair of multispectral images to be processed; an extraction module, used to extract shallow features from the acquired pair of multispectral images using a spectral selection network; a processing module, used to process the shallow features using a feature extraction network to generate a feature vector for each pixel in the image; wherein, the proposed loss function is used for training in the training phase, and descriptors are extracted according to matching needs in the inference phase; wherein, the nearest neighbor algorithm is used to perform coarse matching on existing descriptors, and the RANSAC method is used for filtering to obtain the final matching result.

[0047] In this embodiment 1, the above-mentioned system is used to implement a multispectral image local feature extraction method based on deep learning, including: obtaining a pair of multispectral images to be processed; using a spectral selection network to extract shallow features from the acquired pair of multispectral images; using a feature extraction network to process the shallow features to generate a feature vector for each pixel in the image; wherein, the proposed loss function is used for training in the training phase, and descriptors are extracted according to matching needs in the inference phase; wherein, the nearest neighbor algorithm is used to perform coarse matching on existing descriptors, and the RANSAC method is used for filtering to obtain the final matching result.

[0048] The image is input into the original backbone network to generate an adaptive gate function determined by the input image. The gate function is then multiplied to the filter of the convolutional network, activating only part of the convolution kernel of the network to extract shallow features of the image.

[0049] The shallow features of the extracted image are:

[0050]

[0051] in, represents the feature map output of the ith channel of the ith layer, represents the i-th convolution kernel of the ,th layer, Indicates whether the i-th convolution kernel of the network layer is activated, and φ represents the nonlinear activation function; if not activated, the output is not 0, and if activated, the feature map I l (x) Perform convolution operation.

[0052] The loss function for training multi-spectral local features is constructed based on the information entropy loss function. When the image is input into the neural network, a vector with a length of 128 dimensions is generated for each point in the original image position.

[0053] During the training process, a spectral selection network is used to generate a descriptor with a length of 128 for each pixel in the multispectral image; after introducing more negative samples, the loss function can reduce the problems caused by the traditional triplet loss function and improve the accuracy of multispectral image registration.

[0054] The loss function L is:

[0055]

[0056] Where N is the total number of samples, z i represents the feature representation of the i-th sample, z i+ Indicates that z i Matching feature z i-j Represents negative samples between different graphs, z i-m Indicates that z iAdjacent negative samples in the same image, z i-n Indicates that z i+ For adjacent negative samples in the same image, τ, τ2, and τ3 represent the temperature values ​​of negative samples from different images and negative samples from the same image, respectively. The temperature values ​​are set to make the samples have better discrimination after training.

[0057] Example 2

[0058] like Figure 1 to Figure 2 As shown, this embodiment 2 provides a method for extracting local features of multispectral images based on deep learning, which designs a spectral selection network that adaptively activates neural network channels according to features, can simultaneously accept multiple images from different spectra for training, and designs a contrast loss function containing more negative samples to extract multispectral image features.

[0059] In this embodiment, the feature training process is as follows:

[0060] Step 1: Extract shallow features: Input a pair (two) of multispectral images into the spectral selection network to extract shallow features.

[0061] Step 2: Extract descriptors: Input shallow features into a feature extraction network to generate feature vectors for each pixel in the image. Use the proposed loss function for training in the Sun Lian stage, and extract descriptors according to matching needs in the inference stage. Step 3: Descriptor matching: Use the nearest neighbor algorithm to perform rough matching on existing descriptors, and use the RANSAC method to filter to obtain the final matching results.

[0062] In multimodal image matching, methods such as ReDFeat or CMMNet use pseudo-twin networks to process different modalities separately without sharing parameters. Although this method works well when processing fixed spectral pairs (such as visible light and infrared), it requires independent models and encounters difficulties when processing variable spectral combinations, which increases the computational cost and limits its feasibility in practical applications. To address these limitations, in this embodiment, a spectral gating module (Spectra-Gate Module, SGM), a convolutional structure with conditional activation filters, is introduced. This module can dynamically adjust to adapt to the spectral characteristics of the input image, generate a gating mask to activate specific filters, so that image processing of different spectra no longer requires the use of independent models for each spectral combination. This approach simplifies the model and reduces computational requirements.

[0063] In this embodiment hi, the designed network structure is named SpectralGate. The SpectralGate network is based on a VGG-like network. We first input the image into the original backbone network, generate an adaptive gate function determined by the input image, and then multiply the gate function to the filter of the convolutional network, activating only part of the convolution kernel of the network to extract the shallow features of the image. The corresponding formula is as follows:

[0064]

[0065] In the formula, represents the feature map output of the i-th channel in the first layer, Represents the i-th convolution kernel in layer 1 Indicates whether the i-th convolution kernel of layer 1 of the network is activated, and φ represents the nonlinear activation function. If it is not activated, the output is not 0, and if it is activated, the feature map I l (x) Perform convolution operation.

[0066] The SpectralGate network uses this method to find the corresponding output gate function for networks with different spectra and extract the respective spectral features under different spectra.

[0067] In this embodiment, the design of the descriptor contrast loss function is introduced as follows:

[0068] The traditional triplet loss includes an anchor sample, a negative sample, and a positive sample. The anchor sample and the negative sample are pushed farther away, while the anchor sample and the positive sample are pulled closer. However, the traditional triplet loss fails to take into account the distance between the negative sample and the positive sample, as shown by the red arrow. After optimization, the anchor sample and the negative sample are pushed farther away, while the anchor sample and the positive sample are pulled closer. The distance between the positive sample and the negative sample is also reduced. In the registration task, this phenomenon may cause misalignment. Therefore, in this embodiment, a loss function InfoSGN Loss for training multi-spectral local features is designed. The loss function is constructed based on the information entropy loss function. When the image is input into the neural network, a vector with a length of 128 dimensions is generated for each point in the original image position.

[0069] As shown in formula (2), where L is the loss function, N is the total number of samples, and z i represents the feature representation of the i-th sample, z i+ Indicates that z i Matching feature z i-j Represents negative samples between different graphs, z i-m Indicates that z i Adjacent negative samples in the same image. i-n Indicates that z i+Adjacent negative samples in the same image. τ, τ2, and τ3 represent the temperature values ​​of negative samples in different images and negative samples in the same image, respectively. The temperature values ​​are set to make the samples have better discrimination after training.

[0070] During the training process, the spectral selection network is used to generate a descriptor with a length of 128 for each pixel in the multispectral image. In the reference image and the test image, 4000 corresponding pairs of points are randomly extracted and trained using the method shown in Formula 2.

[0071] After introducing more negative samples, the loss function can reduce the problems caused by the traditional triplet loss function and improve the accuracy of multispectral image registration.

[0072] Example 3

[0073] This embodiment 3 provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the multispectral image local feature extraction method based on deep learning as described above is implemented. The method includes:

[0074] Obtaining a pair of multispectral images to be processed;

[0075] The spectral selection network is used to extract shallow features from the acquired one-to-multispectral images;

[0076] The shallow features are processed by the feature extraction network to generate the feature vector of each pixel in the image; the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching results.

[0077] Example 4

[0078] This embodiment 4 provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the multispectral image local feature extraction method based on deep learning as described above, the method comprising:

[0079] Obtaining a pair of multispectral images to be processed;

[0080] The spectral selection network is used to extract shallow features from the acquired one-to-multispectral images;

[0081] The shallow features are processed by the feature extraction network to generate the feature vector of each pixel in the image; the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching results.

[0082] Example 5

[0083] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the multispectral image local feature extraction method based on deep learning as described above, the method comprising:

[0084] Obtaining a pair of multispectral images to be processed;

[0085] The spectral selection network is used to extract shallow features from the acquired one-to-multispectral images;

[0086] The shallow features are processed by the feature extraction network to generate the feature vector of each pixel in the image; the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching results.

[0087] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0089] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0091] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Technical personnel in the relevant field should understand that various modifications or variations that can be made by those skilled in the art without creative work on the basis of the technical solution disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for extracting local features of multispectral images based on deep learning, characterized in that: include: Obtaining a pair of multispectral images to be processed; The spectral selection network is used to extract shallow features from the acquired one-to-multispectral images; The shallow features are processed by the feature extraction network to generate the feature vector of each pixel in the image; the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching results.

2. The method for extracting local features from multispectral images based on deep learning according to claim 1, characterized in that: The image is input into the original backbone network to generate an adaptive gate function determined by the input image. The gate function is then multiplied to the filter of the convolutional network, activating only part of the convolution kernel of the network to extract shallow features of the image.

3. The method for extracting local features from multispectral images based on deep learning according to claim 2, characterized in that: The shallow features of the extracted image are: in, represents the feature map output of the i-th channel of the l-th layer, represents the i-th convolution kernel of the l-th layer, Indicates whether the i-th convolution kernel of the l-th layer of the network is activated, and φ represents the nonlinear activation function; if it is not activated, the output is not 0, and if it is activated, the feature map I l (x) Perform convolution operation.

4. The method for extracting local features from multispectral images based on deep learning according to claim 1, characterized in that: The loss function for training multi-spectral local features is constructed based on the information entropy loss function. When the image is input into the neural network, a vector with a length of 128 dimensions is generated for each point in the original image position.

5. The method for extracting local features from multispectral images based on deep learning according to claim 4, characterized in that: During the training process, a spectral selection network is used to generate a descriptor with a length of 128 for each pixel in the multispectral image; after introducing more negative samples, the loss function can reduce the problems caused by the traditional triplet loss function and improve the accuracy of multispectral image registration.

6. The method for extracting local features from multispectral images based on deep learning according to claim 5, characterized in that: The loss function L is: Where N is the total number of samples, z i represents the feature representation of the i-th sample, z i+ Indicates that z i Matching feature z i-j Represents negative samples between different graphs, z i-m Indicates that z i Adjacent negative samples in the same image, z i-n Indicates that z i+ For adjacent negative samples in the same image, τ, τ2, and τ3 represent the temperature values ​​of negative samples from different images and negative samples from the same image, respectively. The temperature values ​​are set to make the samples have better discrimination after training.

7. A multispectral image local feature extraction system based on deep learning, characterized in that: include: An acquisition module, used for acquiring a pair of multispectral images to be processed; An extraction module, used for extracting shallow features from the acquired one-to-multi-spectral image using a spectral selection network; The processing module is used to process the shallow features using the feature extraction network to generate the feature vector of each pixel in the image; wherein, the proposed loss function is used for training in the training phase, and the descriptors are extracted according to the matching needs in the inference phase; wherein, the nearest neighbor algorithm is used to perform coarse matching on the existing descriptors, and the RANSAC method is used for filtering to obtain the final matching result.

8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the multispectral image local feature extraction method based on deep learning as described in any one of claims 1 to 6 is implemented.

9. A computer device, characterized in that: It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the multispectral image local feature extraction method based on deep learning as described in any one of claims 1-6.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the multispectral image local feature extraction method based on deep learning as described in any one of claims 1 to 6.

Images