A method and system for detecting small infrared targets
By combining spatial and frequency domain methods, multi-scale dilated contrast convolution and dynamic high-pass filters are used to enhance infrared small target detection, solving the problem of small target recognition in complex backgrounds and achieving high-precision infrared small target detection.
Patent Information
- Application Number
- CN202510062080.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-15
- Publication Date
- 2025-11-14
- Estimated Expiration
- 2045-01-15
AI Technical Summary
Existing infrared small target detection methods have difficulty accurately identifying and separating small targets in complex backgrounds, especially in low signal-to-noise ratio and low frequency backgrounds, which can easily lead to false alarms and missed detections.
A method combining spatial and frequency domains is adopted, which utilizes a multi-scale dilated contrast convolution module to enhance target perception and a dynamic high-pass filter module to remove low-frequency information, thereby enhancing the infrared small target detection performance.
It effectively improves the detection accuracy of small and variable-sized targets, reduces background interference, and enhances the accuracy and reliability of detection.
Smart Images
Figure CN120047665B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of target detection technology, specifically to an infrared small target detection method and system. Background Technology
[0002] Infrared Small Target Detection (IRSTD) aims to identify small targets in infrared images and separate them from complex backgrounds. It has a wide range of applications, including military reconnaissance and traffic monitoring. Compared with general object detection in natural scene images, small target detection in infrared images faces two main challenges: their small and variable size, and their dim appearance and low signal-to-noise ratio. Due to the varying distances between the camera and the target, infrared small targets occupy only a few to tens of pixels, making accurate separation difficult. In addition, there are many slowly changing low-frequency background regions in infrared images, which reduce the visibility and contrast of targets, making them easily occluded. All these obstacles can lead to false alarms or missed detections, making IRSTD a challenging task.
[0003] Current infrared small target detection methods are mainly divided into two categories: traditional methods and deep learning-based methods. Traditional methods, such as filter-based methods, low-rank representation-based methods, and local contrast-based methods, aim to enhance local contrast or highlight target regions through manually designed priors or structural assumptions. However, infrared small target images in real-world scenes are much more complex, with significant variations in target size and cluttered backgrounds, making it difficult for traditional methods to handle such variations using handcrafted features and fixed hyperparameters. In contrast, deep learning-based methods utilize an end-to-end learning paradigm, employing gradient descent algorithms to learn features from a large number of complex scenes, achieving better performance than traditional methods. These methods rely on model design to achieve more efficient feature extraction; however, the inherent ambiguity of small targets often misleads the model, making it susceptible to interference from complex backgrounds, and targets of different scales may lead to false positives.
[0004] In recent years, with the development of deep learning, especially Convolutional Neural Networks (CNNs), end-to-end networks such as CNNs have been able to learn features directly from data without the need for complex manual feature design. Although these deep learning methods have achieved satisfactory performance in most simple scenarios, they still lack comprehensive discrimination and separation capabilities for small, variable-sized, dimly lit targets with low signal-to-noise ratios. In particular, when the background and target have similar frequency components, the model may misidentify the background as the target, leading to false positives and missed detections. Therefore, accurately identifying infrared small targets of different sizes in low-frequency backgrounds and with low signal-to-noise ratios remains a challenging task that has not yet been fully explored. Summary of the Invention
[0005] To address the aforementioned issues, this disclosure proposes an infrared small target detection method and system. Based on two branches—spatial and frequency domains—it utilizes the multi-scale target sensing capability of the spatial domain and the low-frequency information suppression capability of the frequency domain to enhance the performance of infrared small target detection.
[0006] According to some embodiments, the present disclosure adopts the following technical solutions:
[0007] An infrared small target detection method includes:
[0008] Using a spatial domain detection model, multi-scale small target detection is performed on the infrared image to be detected, generating a spatial domain prediction map;
[0009] Based on the spatial domain prediction map, a frequency domain enhancement model is used to perform multi-level high-frequency information enhancement on the frequency domain map corresponding to the infrared image to generate a frequency domain prediction map.
[0010] The spatial domain prediction map and the frequency domain prediction map are fused to obtain the final small target detection map;
[0011] The spatial domain detection model enhances the perception of small infrared targets by using a multi-scale dilated contrast convolution module with multiple parallel dilated contrast convolutions of different kernel sizes. The frequency domain enhancement model uses a dynamic high-pass filter module to calculate low-frequency signal energy in layers and dynamically remove low-frequency information to preserve high-frequency image details.
[0012] According to some embodiments, the present disclosure adopts the following technical solutions:
[0013] An infrared small target detection system, comprising:
[0014] The spatial domain module is configured to: use a spatial domain detection model to perform multi-scale small target detection on the infrared image to be detected and generate a spatial domain prediction map;
[0015] The frequency domain module is configured to: based on the spatial domain prediction map, perform multi-level high-frequency information enhancement on the frequency domain map corresponding to the infrared image through a frequency domain enhancement model to generate a frequency domain prediction map;
[0016] The fusion module is configured to fuse the spatial domain prediction map and the frequency domain prediction map to obtain the final small target detection map.
[0017] The spatial domain detection model enhances the perception of small infrared targets by using a multi-scale dilated contrast convolution module with multiple parallel dilated contrast convolutions of different kernel sizes. The frequency domain enhancement model uses a dynamic high-pass filter module to calculate low-frequency signal energy in layers and dynamically remove low-frequency information to preserve high-frequency image details.
[0018] According to some embodiments, the present disclosure adopts the following technical solutions:
[0019] A computer program product includes a computer program that, when executed by a processor, implements the infrared small target detection method.
[0020] According to some embodiments, the present disclosure adopts the following technical solutions:
[0021] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the infrared small target detection method described above.
[0022] According to some embodiments, the present disclosure adopts the following technical solutions:
[0023] An electronic device includes 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 to enable the electronic device to perform the infrared small target detection method described above.
[0024] Compared with the prior art, the beneficial effects of this disclosure are as follows:
[0025] 1. This invention provides an infrared small target detection method and system, which is based on two branches: spatial domain and frequency domain. It utilizes the multi-scale target perception capability of the spatial domain and the low-frequency information suppression capability of the frequency domain to enhance the performance of infrared small target detection.
[0026] 2. This invention proposes a spatial domain multi-scale dilated contrast convolution module, which features three specially designed parallel dilated contrast convolutions with different kernel sizes. This improves the contrast between the target and the cluttered background, enhancing the perception capability for small and variable-sized targets.
[0027] 3. This invention proposes a frequency domain dynamic high-pass filter module that calculates low-frequency signal energy in layers and dynamically removes specific low-frequency information to preserve high-frequency image details, effectively filtering out slowly changing low-frequency background interference and highlighting small targets. Attached Figure Description
[0028] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.
[0029] Figure 1 This is a model framework diagram proposed in Example 1;
[0030] Figure 2 This is an internal structure diagram of the multi-scale dilated contrast convolution module in Example 1;
[0031] Figure 3 This is an internal structure diagram of the dynamic high-pass filter module in Example 1;
[0032] Figure 4 This is a visualization of the method in Example 1 compared to five existing networks. Detailed Implementation
[0033] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.
[0034] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.
[0035] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0036] Example 1
[0037] One embodiment of this disclosure provides an infrared small target detection method, including:
[0038] Step 1: Using a spatial domain detection model, perform multi-scale small target detection on the infrared image to be detected, and generate a spatial domain prediction map;
[0039] Step 2: Based on the spatial domain prediction map, the frequency domain map corresponding to the infrared image is enhanced with multiple levels of high-frequency information through a frequency domain enhancement model to generate a frequency domain prediction map;
[0040] Step 3: Fuse the spatial domain prediction map and the frequency domain prediction map to obtain the final small target detection map;
[0041] The spatial domain detection model enhances the perception of small infrared targets by using a multi-scale dilated contrast convolution module with multiple parallel dilated contrast convolutions of different kernel sizes. The frequency domain enhancement model uses a dynamic high-pass filter module to calculate low-frequency signal energy in layers and dynamically remove low-frequency information to preserve high-frequency image details.
[0042] As one embodiment, the infrared small target detection method disclosed herein is based on two branches: spatial domain and frequency domain. It utilizes the multi-scale target perception capability of the spatial domain and the low-frequency information suppression capability of the frequency domain to enhance the performance of infrared small target detection, especially to enhance the perception of small and variable-sized targets. The specific implementation process is described in detail below.
[0043] This embodiment designs an infrared small target detection network based on a hybrid domain, such as... Figure 1 As shown, it includes a spatial domain detection model, a frequency domain enhancement model, and a fusion model, where the spatial domain detection model and the frequency domain enhancement model correspond to the spatial domain branch and the frequency domain branch, respectively.
[0044] The spatial domain detection model employs a classic encoder-decoder framework. The encoder uses a specially designed multi-scale dilated contrast convolution (MAC) module as its core component, while the decoder uses ordinary convolutional blocks to progressively upsample and recover target information. The encoder output features are fused with the decoder through skip connections to enhance contextual information capture. The four decoder stages generate multi-level prediction maps, which helps suppress complex background interference in the following frequency domains.
[0045] Each stage of the frequency domain enhancement model integrates a specially designed dynamic high-pass filter (DHPF) module to progressively remove low-frequency information from the image.
[0046] The fusion model combines the final spatial prediction map with the frequency domain prediction map to generate the final infrared small target prediction map.
[0047] Here is a specific example of using an infrared small target detection network for detection, with the following steps:
[0048] S1: Using the multi-scale dilated contrast convolution (MAC) module, feature extraction is performed on the input image in the encoder stage of the spatial domain detection model; and the final spatial prediction map and decoder stage prediction map are generated.
[0049] The spatial domain detection model, such as Figure 1 The spatial domain portion of the diagram is shown, comprising several encoder layers and several decoder layers. The encoder includes a multi-scale dilated contrast convolution module, pooling layers, residual blocks, and upsampling blocks, used for multi-scale feature extraction and providing skip connection inputs to the decoder. The decoder, based on the extracted multi-scale features, obtains prediction maps of different scales through convolution blocks.
[0050] Specifically, the five-layer encoder performs multi-scale feature extraction through the MAC module and provides skip connection inputs for the decoder stage; the four-layer decoder obtains four prediction maps P1, P2, P3, and P4 at different scales through 1×1 convolutional blocks, and obtains the final spatial domain prediction map P0 using the following formula:
[0051] P0 = Conv 1×1 (Concat(P1,BU-2(P2),BU-4(P4),BU-8(P8))),
[0052] Among them, Conv 1×1 (·) is a 1×1 convolutional block, Concat(·) refers to the channel fusion operation, and BU-X(·) refers to upsampling the feature map by X times.
[0053] MAC module, such as Figure 2 As shown, it includes four branches: one direct connection branch and three parallel hole comparison branches. The specific processing steps are as follows:
[0054] S11: Given the input feature map F in Using 1×1 convolutions, the channel dimensions are expanded, and a feature map X is generated:
[0055] X = Conv 1×1 (F in )
[0056] Among them, Conv 1×1 (·) represents a 1×1 convolutional block, designed to expand the channel dimension.
[0057] S12: The feature map X is evenly divided into four groups along the channel dimension, represented as X = [X0, X1, X2, X3].
[0058] X0 is a direct connection and is output immediately, marked as Z0, which is the direct connection branch.
[0059] X1, X2, and X3 are input in parallel to three different dilated contrastive convolution branches, namely ACC-1, ACC-2, and ACC-3, producing contrast features with different receptive fields, denoted as Z1, Z2, and Z3. The specific calculation is as follows:
[0060]
[0061] Wherein, ACC1(·), ACC2(·), and ACC3(·) represent three parallel multi-scale dilated contrast convolution kernels, and the specific calculation method is as follows:
[0062]
[0063] Where (u,v) represents the feature map X i The pixel coordinates within a contrast convolution kernel. Y represents the set of center pixels marked in yellow, and B represents the pixel coordinates within a contrast convolution kernel. Figure 2 The surrounding pixel set (ACC-1, ACC-2, ACC-3) is marked in blue, and n and m represent the total number of pixels in the center or around a contrast kernel.
[0064] S13: Connect all features Z0, Z1, Z2, and Z3 along the channel dimension to obtain the contrast feature Z = [Z0, Z1, Z2, Z3], and perform feature mixing on Z through a 1×1 convolution block.
[0065] S14: Obtain the contrast-enhanced feature map by performing residual block processing on the feature map X and the contrast feature Z after feature mixing on the directly connected branch using the following formula:
[0066] F out =Res_Block(Conv 1×1 (Z)+X),
[0067] Among them, Conv 1×1 (·) refers to a 1×1 convolutional block, and Res_Block(·) is determined by Conv 3×3 It consists of (·), channel attention CA(·), and spatial attention SA(·).
[0068] S2: Using the Dynamic High-Pass Filter (DHPF) module, in the frequency domain enhancement model, the spatial prediction map generated by the spatial domain detection model and the decoder stage prediction map are filtered and low-frequency energy is reduced to obtain the frequency domain feature map.
[0069] The frequency domain enhancement model, such as Figure 1 As shown in the frequency domain section, it consists of several cascaded dynamic high-pass filter (DHPF) modules. The target enhancement feature map output by the last dynamic high-pass filter module serves as the frequency domain prediction map.
[0070] The four Dynamic High-Pass Filter (DHPF) modules have the same structure. Taking the first DHPF module as an example, as follows: Figure 3 As shown in (a) above, the specific processing steps are as follows:
[0071] S21: Generate the frequency domain graph through Fast Fourier Transform.
[0072] First, the predicted image P4 output by the decoder is upsampled by 8 times, and the predicted image P'4 is obtained by passing the sigmoid function. Initial filtering is applied to the predicted image P'4 to effectively reduce background interference and generate the target enhancement image. Then, the target enhancement image is converted into a frequency domain image, i.e., a frequency feature map, using the formula:
[0073]
[0074] Where, IRI∈R H×W×1 and P'4∈R H×W×1 Represents the original infrared image and the predicted image generated by the corresponding decoder level, respectively; ⊙ denotes element-wise multiplication. It refers to the frequency feature map, and FFT(·) refers to the Fast Fourier Transform algorithm.
[0075] S22: Gradually filter out a preset proportion of low-frequency information from the frequency domain graph.
[0076] Figure 3 The frequency characteristic diagram shown in (a) is as follows. This indicates that the frequency decreases and the amplitude increases as the image approaches the center, while the frequency increases and the amplitude decreases as the image moves away from the center. This suggests that the infrared image contains abundant low-frequency background information. Since small target areas and edge areas are both high-frequency information, this embodiment proposes a method to gradually filter out a certain proportion of low-frequency information in order to effectively eliminate low-frequency background and high-frequency edge background information. The specific steps are as follows:
[0077] (1) Calculate the frequency characteristic map Energy:
[0078]
[0079] Where (u,v) represents the pixel coordinates in the frequency feature map. EC represents the amplitude value at pixel (u,v), and EC represents the total frequency domain energy of the image.
[0080] (2) Calculate the dynamic filtering mask radius of the frequency domain map based on the preset energy removal ratio, expressed by the formula:
[0081]
[0082] Among them, E removed λ1 represents the low-frequency energy to be suppressed in the DHPF stage, λ1 refers to the preset energy removal ratio, and the dynamic filter mask radius d is taken as the maximum radius value that satisfies the above equation.
[0083] (3) Based on the radius, the dynamic filtering mask is obtained as follows:
[0084]
[0085] Here, Mask(u,v)∈R H×W×1 This indicates a dynamic filtering mask.
[0086] Since the initial infrared image contains abundant low-frequency information, and its predicted map shows a reduction in low-frequency background as the decoder transitions to a shallower layer, the energy filter ratio is gradually reduced and preset to λ = [λ1, λ2, λ3, λ4] = [0.8, 0.4, 0.2, 0.1].
[0087] (4) The frequency domain map is processed using a dynamic filtering mask to obtain the target enhancement feature map:
[0088]
[0089] Where iFFT(·) represents the inverse fast Fourier transform, This represents the output of the first DHPF module. The frequency domain visualization feature map is shown below. Figure 3 As shown in (d) in the figure, it can be seen that with the different λ in the DHPF modules i The reduction in frequency means a decrease in low-frequency information. Figure 3 The spatial domain prediction plots shown in columns 2-5 of (b)(c) indicate that, with λ i As the frequency decreases, large areas of low-frequency background information and edge background regions of high-frequency information are gradually suppressed.
[0090] Finally, the frequency domain enhancement model, after four DHPF iterations, obtains a prediction map in the frequency domain space.
[0091] S3: Merge the spatial prediction map and the frequency domain prediction map to generate the final infrared small target prediction map.
[0092] Specifically, the final prediction map P is obtained by fusing the prediction map P0 in the spatial domain and the prediction map P0 in the frequency domain. To obtain, expressed by the formula:
[0093]
[0094] Where sig(·) is the Sigmoid activation function, and P represents the infrared small target prediction map, which can effectively suppress the background region.
[0095] An infrared small target detection network consisting of a spatial domain detection model, a frequency domain enhancement model, and a fusion model is trained, and a position-sensitive loss function is used to supervise the training process. The position-sensitive loss function consists of a scale-sensitive loss and a position-sensitive loss.
[0096] Specifically, during the training process, in order to ensure the effective learning and optimization of the model, a deep learning supervision mechanism was adopted to supervise the prediction maps (including the intermediate prediction maps of the spatial domain detection model and the final prediction maps output by the infrared small target detection network) in multiple ways. The performance of infrared small target detection was improved by calculating the loss function of all intermediate prediction maps and the final prediction maps generated by each decoder.
[0097] The loss function used is the Scale and Location Sensitive Loss (SLSLoss) to achieve a comprehensive improvement in the model's detection accuracy. The formula for the Scale and Location Sensitive Loss is:
[0098] L SLS =L S +L L
[0099] Among them, L S and L L These refer to scale-sensitive loss and location-sensitive loss, respectively.
[0100] Scale-sensitive loss L S The definition of is:
[0101]
[0102] Where P and G represent the prediction map and the ground truth map, |·| represents the count of the pixel set, and Var(·) calculates the variance of a given scalar.
[0103] Location-sensitive loss L L The definition of is:
[0104]
[0105] Among them, c p =(x p ,y p ) and c gt =(x gt ,y gt (d) are the center points of the predicted target pixel set in the prediction map P and the ground truth target pixel set in the ground truth map G, respectively. They are calculated by averaging the coordinates of all pixels in each pixel set. Then, we transform the coordinates of these center points to polar coordinates. p ,θ p ) and (d gt,θ gt ) represent c respectively p and c gt The distance and angle.
[0106] Based on the location-sensitive loss function L SLS In this embodiment, a multi-scale ground truth map is constructed to supervise the prediction maps generated at different stages, namely the four intermediate prediction maps P1, P2, P3, and P4 of the spatial domain detection model and the final prediction map P output by the infrared small target detection network. The specific loss design is as follows:
[0107]
[0108] Where G is the truth graph. It is an operation that uses the second parameter as a factor to spatially downsample the first parameter. It is a constructed multi-scale truth map, P i (i∈[1,2,3,4]) refers to the prediction graph obtained from the four decoder stages, and P refers to the final prediction graph.
[0109] This embodiment compares the performance of our method and existing methods (PBT, SCTransNet, MSHNet, RPCANet, and MTUNet) on the same datasets (three international standard datasets: IRSTD-1K, NUAA-SIRST, and NUDT-SIRST). The detection accuracy is evaluated using the metrics IoU, Pd, and Fa. The experimental results are shown in Table 1 and... Figure 4 As shown:
[0110] Table 1 Comparison of Detection Accuracy
[0111]
[0112] In Table 1, among the nine metrics of the three datasets, the method of this embodiment achieved seven best performances, one second-best performance, and one third-best performance. In particular, it achieved 100% target-level Pd accuracy in the NUAA-SIRST dataset. Figure 4 The visualization results show the comparison between the method of this embodiment and five existing networks. Correctly detected targets, missed targets, and falsely detected targets are outlined in red, blue, and yellow boxes, respectively. A magnified view of the targets is shown in the corner of the image. It can be seen that the existing networks (columns 4-8) have obvious false detections and missed detections, resulting in higher Fa. However, the method of this embodiment can accurately identify and segment these small targets. Therefore, this embodiment can effectively enhance the performance of infrared small target detection.
[0113] Example 2
[0114] One embodiment of this disclosure provides an infrared small target detection system, including:
[0115] The spatial domain module is configured to: use a spatial domain detection model to perform multi-scale small target detection on the infrared image to be detected and generate a spatial domain prediction map;
[0116] The frequency domain module is configured to: based on the spatial domain prediction map, perform multi-level high-frequency information enhancement on the frequency domain map corresponding to the infrared image through a frequency domain enhancement model to generate a frequency domain prediction map;
[0117] The fusion module is configured to fuse the spatial domain prediction map and the frequency domain prediction map to obtain the final small target detection map.
[0118] The spatial domain detection model enhances the perception of small infrared targets by using a multi-scale dilated contrast convolution module with multiple parallel dilated contrast convolutions of different kernel sizes. The frequency domain enhancement model uses a dynamic high-pass filter module to calculate low-frequency signal energy in layers and dynamically remove low-frequency information to preserve high-frequency image details.
[0119] Example 3
[0120] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the infrared small target detection method described above.
[0121] Example 4
[0122] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the infrared small target detection method.
[0123] Example 5
[0124] One embodiment of this disclosure 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 to enable the electronic device to perform the infrared small target detection method described above.
[0125] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0126] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0127] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.
Claims
1. A method for detecting small infrared targets, characterized in that, include: Using a spatial domain detection model, multi-scale small target detection is performed on the infrared image to be detected, generating a spatial domain prediction map; Based on the spatial domain prediction map, a frequency domain enhancement model is used to perform multi-level high-frequency information enhancement on the frequency domain map corresponding to the infrared image to generate a frequency domain prediction map. The spatial domain prediction map and the frequency domain prediction map are fused to obtain the final small target detection map; The spatial domain detection model enhances the perception of small infrared targets by using a multi-scale dilated contrast convolution module with multiple parallel dilated contrast convolutions of different kernel sizes; the frequency domain enhancement model uses a dynamic high-pass filter module to calculate low-frequency signal energy in layers and dynamically remove low-frequency information to preserve high-frequency image details. The multi-scale dilated contrast convolution module includes a direct connection branch and several dilated contrast branches. The dilated contrast branches extract contrast features from the feature map using the multi-scale dilated contrast convolution kernel. The specific formula is as follows: in, Representation of feature map Pixel coordinates within, Represents the central pixel set, Represents the surrounding pixel set, and These represent the total number of center pixels and the total number of surrounding pixels, respectively. The frequency domain enhancement model consists of several cascaded dynamic high-pass filter modules. The target enhancement feature map output by the last dynamic high-pass filter module serves as the frequency domain prediction map. The specific operation of the dynamic high-pass filter module is as follows: based on the spatial domain prediction map and the infrared image, a frequency domain map is generated through a fast Fourier transform; the dynamic filtering mask radius of the frequency domain map is calculated according to a preset energy removal ratio; the dynamic filtering mask is obtained based on the radius; and the frequency domain map is processed using the dynamic filtering mask to obtain the target enhancement feature map.
2. The infrared small target detection method as described in claim 1, characterized in that, The spatial domain detection model includes several layers of encoders and several layers of decoders; The encoder includes a multi-scale dilated contrast convolution module, a pooling layer, a residual block, and an upsampling block, used for multi-scale feature extraction and to provide skip connection inputs for the decoder. The decoder, based on the extracted multi-scale features, obtains prediction maps of different scales through convolutional blocks.
3. The infrared small target detection method as described in claim 1, characterized in that, The fusion of the spatial domain prediction map and the frequency domain prediction map is achieved through a fusion model. First, the frequency domain prediction map is activated, then it is multiplied pixel by pixel with the spatial domain prediction map, and finally it is added to the spatial domain prediction map.
4. The infrared small target detection method as described in claim 3, characterized in that, It also includes training the spatial domain detection model, frequency domain enhancement model, and fusion model as a whole, and using a position-sensitive loss function to supervise the training process, wherein the position-sensitive loss function consists of scale-sensitive loss and position-sensitive loss.
5. An infrared small target detection system, characterized in that, include: The spatial domain module is configured to: use a spatial domain detection model to perform multi-scale small target detection on the infrared image to be detected and generate a spatial domain prediction map; The frequency domain module is configured to: based on the spatial domain prediction map, perform multi-level high-frequency information enhancement on the frequency domain map corresponding to the infrared image through a frequency domain enhancement model to generate a frequency domain prediction map; The fusion module is configured to fuse the spatial domain prediction map and the frequency domain prediction map to obtain the final small target detection map. The spatial domain detection model enhances the perception of small infrared targets by using a multi-scale dilated contrast convolution module with multiple parallel dilated contrast convolutions of different kernel sizes; the frequency domain enhancement model uses a dynamic high-pass filter module to calculate low-frequency signal energy in layers and dynamically remove low-frequency information to preserve high-frequency image details. The multi-scale dilated contrast convolution module includes a direct connection branch and several dilated contrast branches. The dilated contrast branches extract contrast features from the feature map using the multi-scale dilated contrast convolution kernel. The specific formula is as follows: in, Representation of feature map Pixel coordinates within, Represents the central pixel set, Represents the surrounding pixel set, and These represent the total number of center pixels and the total number of surrounding pixels, respectively. The frequency domain enhancement model consists of several cascaded dynamic high-pass filter modules. The target enhancement feature map output by the last dynamic high-pass filter module serves as the frequency domain prediction map. The specific operation of the dynamic high-pass filter module is as follows: based on the spatial domain prediction map and the infrared image, a frequency domain map is generated through a fast Fourier transform; the dynamic filtering mask radius of the frequency domain map is calculated according to a preset energy removal ratio; the dynamic filtering mask is obtained based on the radius; and the frequency domain map is processed using the dynamic filtering mask to obtain the target enhancement feature map.
6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the infrared small target detection method according to any one of claims 1-4.
7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement an infrared small target detection method as described in any one of claims 1-4.
8. An electronic device, characterized in that, include: The device includes 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 to enable the electronic device to perform an infrared small target detection method as described in any one of claims 1-4.