Self-adaptive low-illumination visible light and infrared image fusion method for intelligent reconnaissance equipment

Abstract

The invention relates to the technical field of computer vision, and provides a self-adaptive low-illumination visible light and infrared image fusion method for intelligent reconnaissance equipment, which comprises the following steps of: acquiring a visible light image and an infrared image through the intelligent reconnaissance equipment; inputting the visible light image into a trained visible light encoder to obtain deep visible light features; inputting the infrared image into a trained infrared encoder to obtain a deep infrared feature; inputting the deep visible light features and the deep infrared features into a fusion module based on a cross attention mechanism to obtain fusion features; inputting the fused features into a decoder for decoding to obtain gray features; and after the gray features and the visible light color channels are spliced, a fused image with visible light color information is obtained through channel conversion. According to the method, visible light and infrared image fusion in a low-illumination environment is realized, clear and accurate target information is effectively reserved, and intelligent reconnaissance equipment is helped to improve the accuracy and speed of analysis and decision.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and more specifically, to an adaptive low-illumination visible light and infrared image fusion method for intelligent reconnaissance equipment. Background Art

[0002] Intelligent reconnaissance equipment is a detection device that integrates high-performance optoelectronic devices such as visible light cameras, infrared thermal imagers, and laser rangefinders. During use, the collected visible light images are fused with infrared images to provide high-quality fused image data for intelligent reconnaissance equipment under various lighting conditions. However, the infrared and visible light image fusion effect of intelligent reconnaissance equipment under low-light conditions is currently poor, which greatly limits the target detection and tracking performance of intelligent reconnaissance equipment in low-light scenes.

[0003] At present, infrared-visible light image fusion mainly relies on infrared image information to compensate for the loss of scene information in visible light images caused by the decrease in illumination. This results in the inability to express the rich scene information in nighttime visible light images in the fused image, which deviates from the original intention of the image fusion task. Another intuitive solution is to use advanced low-light enhancement algorithms to pre-enhance visible light images, and then merge the source images through fusion methods. However, treating image enhancement and image fusion as separate tasks often leads to incompatibility issues. Specifically, due to the weak light of night scenes, nighttime visible light images will have slight color distortion. Applying low-light enhancement algorithms to them will change the color distribution of the light source, further amplifying the color distortion of the entire image to a certain extent, and ultimately resulting in poor image fusion effects. Summary of the invention

[0004] In order to overcome the defect that the above-mentioned infrared-visible light image fusion method has poor image fusion effect under low-light conditions, the present invention provides an adaptive low-illumination visible light and infrared image fusion method for intelligent reconnaissance equipment.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] An adaptive low-light visible light and infrared image fusion method for intelligent reconnaissance equipment comprises the following steps:

[0007] Collect visible light and infrared images through intelligent reconnaissance equipment;

[0008] Inputting the visible light image into a trained visible light encoder to obtain deep visible light features;

[0009] Inputting the infrared image into a trained infrared encoder to obtain deep infrared features;

[0010] Inputting the deep visible light feature and the deep infrared feature into a fusion module based on a cross attention mechanism to obtain a fusion feature;

[0011] Input the fused features into a decoder for decoding to obtain grayscale features;

[0012] After the grayscale features are spliced ​​with the visible light color channel, a fused image with visible light color information is obtained through channel conversion.

[0013] Furthermore, the present invention also proposes a device, including a memory and a processor, wherein the memory stores computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the processor executes the steps of the anti-interference vehicle detection method proposed in the present invention.

[0014] Furthermore, the present invention also proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement the steps of the anti-interference vehicle detection method proposed in the present invention.

[0015] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0016] The present invention combines a visible light encoder, an infrared encoder, a fusion module based on a cross attention mechanism, and a decoder to form a visible light-infrared image fusion module with a "dual-channel encoding-fusion-decoding" architecture, ensuring that the final fused image has higher contrast and rich texture details;

[0017] The present invention splices the grayscale features fused and decoded by the cross-attention mechanism with the visible light color channel to obtain a fused image with visible light color information, realizes the fusion of visible light and infrared images in low-light environments, effectively retains clear and accurate target information, and helps intelligent reconnaissance equipment to improve the accuracy and speed of analysis and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The figure is a flow chart of a method for adaptively fusing low-illumination visible light and infrared images according to an embodiment of the present invention.

[0019] Figure 2 The figure is a schematic diagram of a visible light-infrared image fusion module according to an embodiment of the present invention.

[0020] Figure 3 FIG. 4 is a structural diagram of a brightness adjustment unit according to an embodiment of the present invention.

[0021] Figure 4 The diagram is a diagram of an architecture of a converged network according to an embodiment of the present invention.

[0022] Figure 5 The figure is a schematic diagram of an image fusion result of an MSRS data set according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0024] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "the" used in the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0025] It should be understood that although the terms first, second, third, etc. may be used in the present invention to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0026] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Example 1

[0028] This embodiment proposes an adaptive low-light visible light and infrared image fusion method for intelligent reconnaissance equipment, such as Figure 1 FIG. 1 is a flow chart of the adaptive low-illumination visible light and infrared image fusion method of this embodiment.

[0029] The adaptive low-light visible light and infrared image fusion method of the intelligent reconnaissance equipment proposed in this embodiment includes the following steps:

[0030] S1. Collect visible light images and infrared images through intelligent reconnaissance equipment;

[0031] S2, inputting the visible light image into a trained visible light encoder to obtain deep visible light features;

[0032] Inputting the infrared image into a trained infrared encoder to obtain deep infrared features;

[0033] S3, inputting the deep visible light feature and the deep infrared feature into a fusion module based on a cross attention mechanism to obtain a fusion feature;

[0034] S4, inputting the fused features into a decoder for decoding to obtain grayscale features;

[0035] S5. After splicing the grayscale features with the visible light color channel, a fused image with visible light color information is obtained through channel conversion.

[0036] This embodiment combines a visible light encoder, an infrared encoder, a fusion module based on a cross attention mechanism, and a decoder to form a visible light-infrared image fusion module of a "dual-channel encoding-fusion-decoding" architecture. Figure 2 The figure shows the architecture of the visible light-infrared image fusion module of this embodiment. The visible light encoder is used to extract deep visible light features, which contain rich texture details in the visible light image, and the infrared encoder is used to extract the contrast information of the target and background in the infrared image, thereby ensuring that the final fused image has higher contrast and rich texture details.

[0037] By extracting deep visible light features and deep infrared features from the image, decoding them after fusion based on the cross-attention mechanism, and splicing them with the visible light color channel, a fused image with visible light color information is obtained, realizing the fusion of visible light and infrared images in low-light environments, effectively retaining clear and accurate target information, and helping intelligent reconnaissance equipment to improve the accuracy and speed of analysis and decision-making.

[0038] In an optional embodiment, the visible light encoder includes a plurality of encoding layers composed of a brightness adjustment unit (BrightnessAdjustment Unit, BAU) and a LeakyReLU activation function. The brightness adjustment unit includes a 3×3 convolution layer and a 1×1 convolution layer. After the input visible light features are subjected to 3×3 convolution operations and 1×1 convolution operations respectively, the extracted features are element-wise multiplied to obtain visible light features.

[0039] For example, Figure 3 The figure shows the architecture of the brightness adjustment unit of this embodiment. In the brightness adjustment unit, the input features are subjected to 1×1 and 3×3 convolution operations with a step size of 1, respectively. When the convolution kernel is 1×1, no padding is performed, and when the convolution kernel is 3×3, the mapping padding is 1; then the features obtained from the two branches are element-wise multiplied.

[0040] In this example, by setting the brightness adjustment unit BAU, it is possible to extract deep features containing rich visible light image texture details from low-illuminance visible light images collected under bad weather or nighttime conditions.

[0041] Exemplarily, the visible light encoder includes 4 groups of encoding layers consisting of brightness adjustment units and LeakyReLU activation functions.

[0042] Furthermore, in step S2, the visible light image is input into a trained visible light encoder, comprising the following steps:

[0043] Perform a balanced histogram operation on the collected visible light image, enhance the overall pixels of the original low-light visible light image, and obtain a visible light color channel CrCb and original grayscale features. Enhanced feature information;

[0044] The original grayscale features The visible light encoder is input and processed by four coding layers consisting of brightness adjustment units and LeakyReLU activation functions to extract deep features F containing rich visible light image texture details. vi .

[0045] In an optional embodiment, the infrared encoder includes several groups of encoding layers consisting of 3×3 convolutional layers and LeakyReLU activation functions.

[0046] Exemplarily, the infrared encoder includes 4 sets of encoding layers consisting of 3×3 convolutional layers and LeakyReLU activation functions, wherein the step size of the convolutional layer is 1 and the mapping padding is 1.

[0047] In this embodiment, the collected infrared image is preprocessed to obtain infrared image information I ir And as the input of the infrared encoder, after being processed by the encoder, the deep infrared feature F rich in contrast information can be extracted ir .

[0048] In an optional embodiment, the fusion network includes a feature concatenation layer, a pooling layer and a cross-attention fusion layer.

[0049] The feature concatenation layer is used to concatenate the input deep visible light features F vi and deep infrared features F ir To splice;

[0050] The pooling layer is used to perform global average pooling (GAP) on the splicing features output by the feature splicing layer to obtain a one-dimensional vector representation of the splicing features;

[0051] The cross-attention fusion layer includes a first branch and a second branch; wherein the first branch performs a convolution operation and a sigmoid function activation on the input splicing feature to obtain a first activation vector v1, and combines it with the deep visible light feature F vi Multiply them to obtain the first screening feature;

[0052] The second branch performs convolution operation and sigmoid function activation on the input splicing feature to obtain a second activation vector v2, and adds it to the deep infrared feature F ir Multiply them to obtain the second screening feature;

[0053] The first screening feature and the second screening feature are spliced ​​to obtain a fusion feature

[0054] For example, Figure 4 FIG. 1 is a schematic diagram of the structure of the fusion network of this embodiment.

[0055] Among them, the convolution operation is used to extract mixed features and reduce the channel dimension, and the sigmoid function is used to activate and filter out the corresponding key feature vectors. The cross-attention fusion operation uses channel attention to enhance the deep features of visible light and infrared by multiplying the corresponding elements of the two channels, and then splicing and fusing them, which helps to enhance the representation ability and texture details of the fused features.

[0056] In an optional embodiment, the decoder includes several groups of decoding layers consisting of 3×3 convolutional layers and LeakyReLU activation functions.

[0057] Exemplarily, the decoder includes 4 sets of decoding layers consisting of 3×3 convolutional layers and LeakyReLU activation functions, wherein the convolutional layer step size is 1 and the mapping padding is 1.

[0058] In this embodiment, the infrared encoder increases the number of image information channels from 1 to 128, while the decoder reduces the number of image information channels from 128 to 1. Therefore, the decoder can not only extract deep features, but also reduce the channel dimension to restore the image information to the dimension of a normal image.

[0059] Furthermore, in an optional embodiment, the method further comprises the following steps:

[0060] Obtain a visible light image data set and an infrared image data set and perform preprocessing to obtain a training data set;

[0061] Initializing all weights and bias items in the visible light infrared image fusion module composed of the visible light encoder and the infrared encoder;

[0062] Construct a loss function that combines scene loss, perceptual loss, and gradient loss;

[0063] The visible light infrared image fusion module is optimized and trained based on the AdamW optimizer and the loss function, and the parameters of the visible light infrared image fusion module are saved.

[0064] Exemplarily, in this embodiment, the MSRS dataset is selected as the training dataset, wherein the images cover various scenes such as nature, city, and indoor.

[0065] Exemplarily, in this embodiment, the Kaiming initialization method is used to initialize all weights and bias items in the network.

[0066] Exemplarily, in this embodiment, the AdamW optimizer is selected for model optimization. Among them, the hyperparameters in the AdamW optimizer are set according to experience, such as momentum factors β1 and β2, initial learning rate ∈, weight decay coefficient λ and decay rate γ. The Bayesian hyperparameter optimization method is adopted, and the weighted combination of average gradient (AG), mutual information (MI) and entropy (EN) used to evaluate the fusion effect is used as the evaluation index. By repeatedly training the model, the model hyperparameters are continuously optimized according to the obtained evaluation index results, and finally a set of relatively optimal hyperparameters is obtained. The optimized hyperparameters are used to guide the training process of the model to ensure the optimization of model performance.

[0067] Furthermore, in an optional embodiment, preprocessing the visible light image dataset and the infrared image dataset comprises the following steps:

[0068] Crop all images to a uniform size;

[0069] Divide the image pixel value by 255 so that the pixel value range is normalized to [0, 1];

[0070] Perform histogram equalization operation on visible light images;

[0071] All images are rotated, flipped, and randomly cropped.

[0072] For example, all images are uniformly cropped to a size of 256×256.

[0073] Furthermore, in an optional embodiment, in the loss function of constructing the joint scene loss, perceptual loss and gradient loss, the scene loss includes the scene loss in the encoding stage. and scene loss in the feature fusion stage Its expression is:

[0074]

[0075] Among them, α1 and α2 are hyperparameters used to adjust the order of magnitude, L1(·) is the L1 norm loss function, that is, the absolute value deviation of the corresponding position element in the vector; F vi Represents the visible light characteristics, F ir Indicates infrared characteristics; Represents grayscale features; Represents fusion features; I ir Indicates the infrared image information extracted from the input infrared image; max(·,·) is used to select the maximum value from the pixel points;

[0076] The perceptual loss l perce Calculated by the trained VGG19 network; its expression is:

[0077]

[0078] Among them, L vgg (·) represents the VGG19 network, which is used to calculate the loss value between the generated features and the target features;

[0079] The gradient loss l grad Including the gradient loss of the encoding stage and the gradient loss in the feature fusion stage Its expression is:

[0080]

[0081] Among them, β1 and β2 are hyperparameters; is the Sobel operator;

[0082] Then, the loss function l join The expression is:

[0083] l join =λ1·l scene +λ2·l perce +λ3·l grad

[0084] Among them, λ1, λ2, and λ3 are hyperparameters used to adjust the scene loss, perceptual loss, and gradient loss to the same order of magnitude.

[0085] Exemplarily, the training of the visible light infrared image fusion module is divided into two stages, the first stage is the encoding stage, and the second stage is the fusion network and decoder stage. The two training stages are based on the scene loss l scene and the perceptual loss l perce The features generated by the encoder are adaptively adjusted to normal brightness through the gradient loss l grad The edge information is extracted, forcing the edge texture to approach the original image, so that the generated image has more texture information.

[0086] In this embodiment, the loss function is constructed by combining scene loss, perceptual loss and gradient loss, which can guide the model to obtain reasonable lighting components, rich scene content and texture details from the fused features.

[0087] As an example, in order to intuitively observe the fusion performance of different algorithms on the MSRS dataset, this embodiment selects a pair of infrared and visible light images, and the image fusion results are as follows: Figure 5 shown.

[0088] Among them, due to the illumination degradation problem of night images, an excellent night fusion algorithm should gather meaningful information from the source images and provide a bright scene with high contrast. Figure 5 It can be seen that in the image fusion effect based on the GANMcC and TarDAL algorithms, the entire scene is submerged in darkness; the fusion results based on algorithms such as Gan-FM, SwinFusion, UMF-CMGR and DIVFusion are relatively blurred and a large amount of edge detail information is lost. Although SeAFusion, STDFusion and SuperFuse have more texture information, the brightness of the entire scene is low. Obviously, the existing image fusion methods cannot provide visual experience that meets the user's needs. The image fusion results of this embodiment can obtain bright scenes and rich textures even in the dark. It is worth emphasizing that none of the nine SOTA methods can display rich details in dark areas, while this embodiment can easily capture texture information without considering dark areas.

[0089] Furthermore, in terms of quantitative evaluation, this embodiment selects six indicators, namely, average gradient (AG), mutual information (MI), entropy (EN), spatial frequency (SF), standard deviation (SD), and visual information fidelity (VIF), to objectively evaluate the fusion effect. Among them, AG reflects the richness of image texture information, MI reflects the amount of information contained in each other, EN evaluates the amount of information contained in the fused image from the perspective of information theory, SF reflects the spatial frequency information contained in the fused image, SD statistically represents the distribution and contrast of the fused image, and VIF measures the fidelity of information from the perspective of human visual perception. The larger the values ​​of the above six indicators, the better the fusion effect. This embodiment selects 30 pairs of images from the MSRS data set to objectively evaluate the SD, MI, VIF, AG, EN and SF indicators, and the fusion result indicator values ​​are shown in Table 1 below.

[0090] Table 1 Six index values ​​of the fusion results of 10 methods on the MSRS dataset

[0091]

[0092] As can be seen from the above table, the method proposed in this embodiment is compared with other existing algorithms. It can be seen that the values ​​of this embodiment in the six indicators are ranked in the top 2. The best results of MI, VIF and SF respectively mean that the results of the method of this embodiment have richer texture details, more scene information and better visual perception. In addition, this embodiment also performs well in SD, AG and EN indicators, which shows that with the help of the loss function and brightness adjustment module, the fusion result of this embodiment is more consistent with the source image and more natural.

[0093] This embodiment significantly improves the fusion effect of visible light images and infrared images at night or in low-light environments, and promotes the development of advanced visual tasks such as target detection and tracking. This means that intelligent reconnaissance equipment can still provide clear and accurate target information at night or in low-light scenes, improving the efficiency of all-weather monitoring.

[0094] Example 2

[0095] This embodiment proposes a device including a memory and a processor, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the adaptive low-illumination visible light and infrared image fusion method proposed in Example 1.

[0096] Example 3

[0097] This embodiment proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the adaptive low-illumination visible light and infrared image fusion method proposed in Embodiment 1.

[0098] Exemplarily, the storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0099] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.

[0100] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the adaptive low-illumination visible light and infrared image fusion method described in Example 1.

[0101] The terms in the drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;

[0102] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An adaptive low-light visible light and infrared image fusion method for intelligent reconnaissance equipment, characterized in that: The following steps are involved: Collect visible light and infrared images through intelligent reconnaissance equipment; Inputting the visible light image into a trained visible light encoder to obtain deep visible light features; Inputting the infrared image into a trained infrared encoder to obtain deep infrared features; Inputting the deep visible light feature and the deep infrared feature into a fusion module based on a cross attention mechanism to obtain a fusion feature; Input the fused features into a decoder for decoding to obtain grayscale features; After the grayscale features are spliced ​​with the visible light color channel, a fused image with visible light color information is obtained through channel conversion.

2. The adaptive low-illumination visible light and infrared image fusion method according to claim 1, characterized in that: The visible light encoder includes several groups of encoding layers consisting of brightness adjustment units and LeakyReLU activation functions; The brightness adjustment unit includes a 3×3 convolution layer and a 1×1 convolution layer. After the input visible light features are subjected to 3×3 convolution operations and 1×1 convolution operations respectively, the extracted features are element-wise multiplied to obtain the visible light features.

3. The adaptive low-illumination visible light and infrared image fusion method according to claim 1, characterized in that: The infrared encoder includes several groups of encoding layers consisting of 3×3 convolutional layers and LeakyReLU activation functions.

4. The adaptive low-illumination visible light and infrared image fusion method according to claim 1, characterized in that: The fusion network includes a feature concatenation layer, a pooling layer and a cross attention fusion layer; wherein: The feature stitching layer is used to stitch the input deep visible light features and deep infrared features; The pooling layer is used to perform maximum average pooling on the spliced ​​features output by the feature splicing layer to obtain a one-dimensional vector representation of the spliced ​​features; The cross attention fusion layer includes a first branch and a second branch; wherein the first branch performs a convolution operation and a sigmoid function activation on the input splicing feature to obtain a first activation vector, and multiplies it with the deep visible light feature to obtain a first screening feature; The second branch performs a convolution operation and a sigmoid function activation on the input splicing feature to obtain a second activation vector, and multiplies the second activation vector with the deep infrared feature to obtain a second screening feature; The first screening feature and the second screening feature are concatenated to obtain a fusion feature.

5. The adaptive low-illumination visible light and infrared image fusion method according to claim 1, characterized in that: The decoder includes several groups of decoding layers consisting of 3×3 convolutional layers and LeakyReLU activation functions.

6. The adaptive low-illumination visible light and infrared image fusion method according to any one of claims 1 to 5, characterized in that: The method further comprises the following steps: Obtain a visible light image data set and an infrared image data set and perform preprocessing to obtain a training data set; Initializing all weights and bias items in the visible light infrared image fusion module composed of the visible light encoder and the infrared encoder; Construct a loss function that combines scene loss, perceptual loss, and gradient loss; The visible light infrared image fusion module is optimized and trained based on the AdamW optimizer and the loss function, and the parameters of the visible light infrared image fusion module are saved.

7. The adaptive low-illumination visible light and infrared image fusion method according to claim 6, characterized in that: Preprocessing the visible light image dataset and the infrared image dataset comprises the following steps: Crop all images to a uniform size; Divide the image pixel value by 255 so that the pixel value range is normalized to [0, 1]; Perform histogram equalization operation on visible light images; All images are rotated, flipped, and randomly cropped.

8. The adaptive low-illumination visible light and infrared image fusion method according to claim 6, characterized in that: In the loss function of constructing the joint scene loss, perceptual loss and gradient loss, the scene loss includes the scene loss in the encoding stage and scene loss in the feature fusion stage Its expression is: Among them, α1 and α2 are hyperparameters used to adjust the order of magnitude, L1(·) is the L1 norm loss function, that is, the absolute value deviation of the corresponding position element in the vector; F vi Represents the visible light characteristics, F ir Indicates infrared characteristics; Represents grayscale features; Represents fusion features; I ir Indicates the infrared image information extracted from the input infrared image; max(·,·) is used to select the maximum value from the pixel points; The perceived loss Calculated by the trained VGG19 network; its expression is: Among them, L vgg (·) represents the VGG19 network, which is used to calculate the loss value between the generated features and the target features; The gradient loss Including the gradient loss in the encoding stage and the gradient loss in the feature fusion stage Its expression is: Among them, β1 and β2 are hyperparameters; is the Sobel operator; Then, the loss function The expression is: Among them, λ1, λ2, and λ3 are hyperparameters used to adjust the scene loss, perceptual loss, and gradient loss to the same order of magnitude.

9. A device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the adaptive low-illumination visible light and infrared image fusion method according to any one of claims 1 to 8.

10. A storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, all or part of the steps of the adaptive low-illumination visible light and infrared image fusion method according to any one of claims 1 to 8 are implemented.

Images