An improved infrared small target detection method, system and device

Abstract

The present invention relates to the field of infrared image detection technology, and provides an improved infrared small target detection method, system and device. An image is acquired and preprocessed to obtain a preprocessed image; a network model is established based on the YOLOv9 framework, and after the preprocessed image is input into the network model, it is processed by a backbone network and a multi-scale feature fusion network in sequence, and is output after being inspected and qualified by a detection head module; the backbone network includes an SD_BS module, and the multi-scale feature fusion network includes a GSELAN module. The present invention solves the technical problem of the loss of shallow information caused by deep convolution by introducing the SD_BS module and the GSELAN module, thereby alleviating the loss of shallow information caused by deep convolution, while reducing the model complexity and the amount of calculation, enhancing the nonlinear expression ability of features, and more efficiently capturing and utilizing the information in the features.

Description

Technical Field

[0001] The present invention relates to the field of infrared image detection technology, and in particular to an improved infrared small target detection method, system and device. Background Art

[0002] Compared with visible light images, infrared images have problems such as low resolution, low contrast, lack of color information, blurred edges, noise interference, and low signal-to-noise ratio. There are a large number of small targets in infrared images. These small targets are usually smaller than 30×30 pixels, occupy a small number of pixels, and are easily submerged by the background. Traditional target detection methods use sliding window technology to move windows on the image and classify each window to detect objects. This method relies on manually designed features and is easily interfered by complex background noise, resulting in poor model robustness, low accuracy, lack of adaptability, and other problems, thereby affecting the accuracy of detection. Deep learning-based methods are widely used in infrared target recognition. Target detection algorithms based on deep learning are divided into one-stage and two-stage detection models. One-stage target detection algorithms directly extract features through the network to further predict the category and location of objects. Representative algorithms include YOLO (You Only Look Once) and SSD (Single Shot MultiBoxDetector). The two-stage target detection process is divided into two steps: first, extract the region of interest of the image, and then classify and locate the target through a classification network. Representative algorithms include the R-CNN (Region-based Convolutional Neural Networks) series and SPP-Net (Spatial Pyramid Pooling Networks) series.

[0003] The YOLO series of models are characterized by high detection speed, but small infrared targets usually have a low signal-to-noise ratio and occupy a small number of pixels, lacking obvious texture and shape features. In deep networks, small targets are easily submerged by complex backgrounds. At the same time, in deep networks, the stacking of multiple layers of convolution will cause the loss of small target features, thereby reducing the detection accuracy of small targets.

[0004] Two-stage object detection algorithms, such as Faster R-CNN, usually need to generate candidate regions first, and then classify and regress these regions. This process is relatively complex, resulting in slow detection speed and difficulty in meeting real-time requirements. And although the two-stage object detection algorithm has high accuracy when detecting large objects, it has poor detection effect for small objects with less feature information. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the related art. To this end, the present invention provides an improved infrared small target detection method, system and device to achieve fast and accurate detection of infrared small targets.

[0006] The present invention provides an improved infrared small target detection method, comprising:

[0007] S1: Acquire and preprocess an image to obtain a preprocessed image;

[0008] S2: Based on the YOLOv9 framework, an infrared small target detection network model is established. The infrared small target detection network model includes a backbone network, a multi-scale feature fusion network, and a detection head module. The backbone network includes an SD_BS module, the multi-scale feature fusion network includes a GSELAN module, and the steps of the SD_BS module are:

[0009] S11: For sizes Input image, input features ,Will After the transformation is split by step size in the size dimension, the slices are divided into a series of sub-features. The formula is:

[0010]

[0011] ...,

[0012] ...,

[0013]

[0014] in, is the number of pixels of the length and width of the input image, is the number of channels, Indicates that the input feature is in the long dimension with the Line starts to So far, in the wide dimension Column starts to So far, Extract pixels for the step size, according to The obtained sub-features are , ;

[0015] S12: Splice the sub-features according to the dimension of the channel to obtain a spliced ​​feature ;

[0016] S13: for the splicing features Perform BN batch normalization to obtain normalized splicing features;

[0017] S14: Use SiLU activation function to perform nonlinearity on the normalized splicing feature to obtain output features ,in ,

[0018] S3: Setting training parameters, using the preprocessed images to train the infrared small target detection network model, and obtaining an infrared small target detection training model;

[0019] S4: Input the detection image to be detected into the infrared small target detection training model to obtain the detection result.

[0020] According to an improved infrared small target detection method provided by the present invention, the GSELAN module replaces the RepNCSPELAN4 module of the multi-scale feature fusion network.

[0021] According to an improved infrared small target detection method provided by the present invention, the structure of the GSELAN module includes:

[0022] S21: Input module inputs data;

[0023] S22: Convolve the module input data with the first module hyperparameter to obtain a first intermediate value;

[0024] S22: performing channel splitting on the first intermediate value to obtain a second intermediate value;

[0025] S23: input the second intermediate value and the second module hyperparameter into the GSCSP_1 module, and perform self-convolution to obtain a third intermediate value;

[0026] S24 inputs the third intermediate value and the second module hyperparameter into the GSCSP_1 module, and performs self-convolution to obtain a fourth intermediate value;

[0027] S25: performing channel splicing on the first intermediate value, the second intermediate value, the third intermediate value, and the fourth intermediate value to obtain a spliced ​​intermediate value;

[0028] S26: Performing deconvolution on the concatenated intermediate value to obtain output data.

[0029] According to an improved infrared small target detection method provided by the present invention, the steps of the GSCSP_1 module include:

[0030] S201: Convolve the input data of the GSCSP_1 module to obtain the intermediate value of the first module;

[0031] S202: Input the input data of the GSCSP_1 module into the GSBottleneck module to obtain the intermediate value of the second module;

[0032] S203: Channel-joining the intermediate value of the first module and the intermediate value of the second module, and then performing self-convolution to obtain output data of the GSCSP_1 module.

[0033] According to an improved infrared small target detection method provided by the present invention, the steps of the GSBottleneck module include:

[0034] S2001: Convolve the input data of the GSBottleneck module to obtain the first convolution result;

[0035] S2002: performing DSC (depth separable convolution) on the first convolution result to obtain a second convolution result;

[0036] S2003: concatenating the first convolution result and the second convolution result, and then performing shuffling processing to obtain a third convolution result;

[0037] S2004: Add the first convolution result and the third convolution result to obtain the output result of the GSBottleneck module.

[0038] According to an improved infrared small target detection method provided by the present invention, the inspection of the detection head module uses three scale feature inspections, including: a large-scale inspection greater than or equal to 80 pixels; a medium-scale inspection greater than or equal to 40 pixels and less than 80 pixels; and a small-scale inspection greater than or equal to 20 pixels and less than 40 pixels.

[0039] The present invention also provides an improved infrared small target detection system, comprising:

[0040] Image preprocessing module: acquire and preprocess the image to obtain the preprocessed image;

[0041] Network model building module: Based on the YOLOv9 framework, an infrared small target detection network model is established. The infrared small target detection network model includes a backbone network, a multi-scale feature fusion network, and a detection head module; the backbone network includes an SD_BS module, and the multi-scale feature fusion network includes a GSELAN module;

[0042] Model training module: setting training parameters, using the preprocessed images to train the infrared small target detection network model, and obtaining an infrared small target detection training model;

[0043] Image detection module: Input the detection image to be detected into the infrared small target detection training model to obtain the detection result.

[0044] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned improved infrared small target detection methods are implemented.

[0045] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0046] The present invention provides an improved infrared small target detection method, system and device, which adopts the SD_BS module in the backbone network to alleviate the loss of shallow information caused by deep convolution, while reducing the model complexity and calculation amount, and enhancing the nonlinear expression ability of features. Combining the programmable gradient information and auxiliary reversible branches of the YOLOv9 model, the shallow gradient can be better updated and the model parameter utilization rate can be improved. The present invention proposes a GSELAN module, and based on this design, an improved neck multi-scale feature fusion network GSELANPANet is designed, which can more efficiently capture and utilize semantic information in features. The network model designed by the present invention can detect infrared small targets more quickly and efficiently.

[0047] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 It is a schematic flow chart of an improved infrared small target detection method provided by the present invention;

[0050] Figure 2 A schematic diagram of the process flow of the SD_BS module provided by the present invention;

[0051] Figure 3 It is a structural block diagram of an improved infrared small target detection device provided by the present invention;

[0052] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention.

[0053] Reference numerals:

[0054] 101. Image preprocessing module; 102. Network model building module; 103. Model training module; 104. Image detection module; 810. Processor; 820. Communication interface; 830. Memory; 840. Communication bus. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme in the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0056] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. 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 embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0057] Example

[0058] Combine the following Figures 1 to 4 Description of the Invention

[0059] like Figure 1 As shown, Figure 1 An improved infrared small target detection method is provided, and the steps are as follows:

[0060] S1: Acquire and preprocess an image to obtain a preprocessed image;

[0061] S2: An infrared small target detection network model is established based on the YOLOv9 framework. The infrared small target detection network model includes a backbone network, a multi-scale feature fusion network, and a detection head module. The backbone network includes an SD_BS module, and the multi-scale feature fusion network includes a GSELAN module.

[0062] S3: Setting training parameters, using the preprocessed images to train the infrared small target detection network model, and obtaining an infrared small target detection training model;

[0063] S4: Input the detection image to be detected into the infrared small target detection training model to obtain the detection result.

[0064] The SD_BS module is used to replace the changeable kernel convolution layer in the backbone network.

[0065] Specifically, Figure 2 As shown, the steps of the SD_BS module are:

[0066] S11: For sizes Input image, input features ,Will After the transformation is split by step size in the size dimension, the slices are divided into a series of sub-features. The formula is:

[0067]

[0068] ...,

[0069] ...,

[0070]

[0071] in, is the number of pixels of the length and width of the input image, is the number of channels, Indicates that the input features are in the long dimension with the Line starts to So far, in the wide dimension Column starts to So far, Extract pixels for the step size, according to The obtained sub-features are , In the embodiment of the present invention , after slicing, we get , , , Four sub-characteristics.

[0072] S12: Splice the sub-features according to the dimension of the channel to obtain a spliced ​​feature ; At this point, the size of the splicing feature becomes the input feature of , and the number of channels is expanded to the input feature of times, the number of pixels of the length and width of the input image in the embodiment of the present invention becomes , the number of channels becomes .

[0073] S13: for the splicing features Perform BN batch normalization to obtain normalized splicing features;

[0074] S14: Use SiLU activation function to perform nonlinearity on the normalized splicing feature to obtain output features ,in The output features obtained by BN batch normalization and SiLU activation function nonlinearization Easier to train and enhance input features The output feature in the embodiment of the present invention is .

[0075] In order to make full use of the feature information of small and weak targets, the embodiment of the present invention proposes an SD_BS module to replace the ordinary convolution module, thereby ensuring the ability of the convolutional neural network to extract target features and making the model more lightweight. Specifically, the module first slices the original features, then concatenates them along the channel dimension, and then adds batch normalization and SiLU (Sigmoid Linear Unit) activation functions to normalize the data and increase the nonlinearity of the features.

[0076] As the network deepens, the spatial information of the input features of the backbone network is gradually transferred to the channel. Each reduction in spatial dimension and increase in channel dimension will result in the loss of some semantic information. Using only DSC can lighten the network to a certain extent, but it will inevitably cause a loss of precision. Therefore, the embodiment of the present invention proposes a GSBottleneck module to solve the problem of deep separation between DSC output feature channels; it enables the local feature information of different channels of Conv (standard convolution) and DSC to interact evenly. And it can achieve more fully retaining effective features with lower computational complexity, so that the model can capture and learn more semantic information in the feature map, which can effectively reduce the complexity of the model and speed up the model reasoning speed.

[0077] The GSELAN module in the embodiment of the present invention replaces the RepNCSPELAN4 module of the multi-scale feature fusion network.

[0078] Specifically, in the embodiment of the present invention, the steps of the GSELAN module include:

[0079] S21: Input module inputs data;

[0080] S22: Convolve the module input data with the first module hyperparameter to obtain a first intermediate value;

[0081] S22: performing channel splitting on the first intermediate value to obtain a second intermediate value;

[0082] S23: input the second intermediate value and the second module hyperparameter into the GSCSP_1 module, and perform self-convolution to obtain a third intermediate value;

[0083] S24 inputs the third intermediate value and the second module hyperparameter into the GSCSP_1 module, and performs self-convolution to obtain a fourth intermediate value;

[0084] S25: performing channel splicing on the first intermediate value, the second intermediate value, the third intermediate value, and the fourth intermediate value to obtain a spliced ​​intermediate value;

[0085] S26: Performing deconvolution on the concatenated intermediate value to obtain output data.

[0086] The steps of the GSCSP_1 module include:

[0087] S201: Convolve the input data of the GSCSP_1 module to obtain the intermediate value of the first module;

[0088] S202: Input the input data of the GSCSP_1 module into the GSBottleneck module to obtain the intermediate value of the second module;

[0089] S203: Channel-joining the intermediate value of the first module and the intermediate value of the second module, and then performing self-convolution to obtain output data of the GSCSP_1 module.

[0090] The steps of the GSBottleneck module include:

[0091] S2001: Convolve the input data of the GSBottleneck module to obtain the first convolution result;

[0092] S2002: performing DSC on the first convolution result to obtain a second convolution result;

[0093] S2003: concatenating the first convolution result and the second convolution result, and then performing shuffling processing to obtain a third convolution result;

[0094] S2004: Add the first convolution result and the third convolution result to obtain the output result of the GSBottleneck module.

[0095] Specifically, the embodiment of the present invention uses the detection head module of the baseline model YOLOv9 model, including: a large-scale test greater than 80 pixels; a medium-scale test greater than or equal to 40 pixels and less than 80 pixels; and a small-scale test greater than or equal to 20 pixels and less than 40 pixels.

[0096] The embodiments of the present invention use two widely used indicators, mAP (Mean Average Precision) and GFLOPS (Giga Floating-point Operations Per Second), for evaluation.

[0097] The calculation formula of mAP is:

[0098]

[0099]

[0100]

[0101] in, For accuracy, is the recall rate, is the number of categories, is the average precision of a single category, is a true positive, For false positive, A false negative is the category ordinal number, .

[0102] In the embodiment of the present invention, the data set is an infrared aerial photography data set. The training infrared aerial photography data set contains 11045 pictures with a resolution of 640×512, including six categories: person, car, bus, cyclist, bike, and truck. The present invention uses 7731 pictures as a training set, 2319 pictures as a validation set, and 995 pictures as a test set. The training uses mAP as the accuracy evaluation indicator.

[0103] The whole model is trained based on the Pytorch deep learning framework, and the model performance is verified using the validation set. The experiments are all trained and tested on the graphics card, and the experimental parameter settings are shown in Table 1.

[0104] Table 1 Experimental parameter settings

[0105]

[0106] Set training parameters, modify the network model configuration file, change the training set address, fine-tune the model parameters and train multiple times to obtain the best detection results. The present invention compares the detection effect of the improved model with the mainstream target detection model. The experimental results are shown in Table 2. The model proposed by the present invention takes into account the requirements of speed and accuracy. Compared with the mainstream detection model, the GFLOPS (representing the computing requirements of the model, the smaller the value, the better the real-time performance of the model) parameter is smaller, and the mAP@0.5 (the mAP value when the intersection-over-union ratio is set to 0.5) reaches the highest. The effect of the present invention compared with the existing detection model is verified by experiments. As shown in Table 2:

[0107] Table 2 Comparative experimental results

[0108]

[0109] Among them, YOLOv5m, YOLOv9m, YOLOv10m, and Faster-RCNN are other model names.

[0110] It can be concluded from the above table that the network model proposed in the present invention has the best detection effect and effectively reduces missed detection and false detection of small targets.

[0111] like Figure 3 As shown, the present invention provides an improved infrared small target detection system, comprising the following modules:

[0112] Image preprocessing module 101: acquiring and preprocessing an image to obtain a preprocessed image;

[0113] Network model building module 102: establishing an infrared small target detection network model based on the YOLOv9 framework, the infrared small target detection network model includes a backbone network, a multi-scale feature fusion network, and a detection head module; the backbone network includes an SD_BS module, and the multi-scale feature fusion network includes a GSELAN module;

[0114] Model training module 103: setting training parameters, using the preprocessed images to train the infrared small target detection network model, and obtaining an infrared small target detection training model;

[0115] Image detection module 104: inputs the detection image to be detected into the infrared small target detection training model to obtain the detection result.

[0116] An improved infrared small target detection device provided by the present invention is described below. The improved infrared small target detection device described below and the improved infrared small target detection method described above can be referred to each other.

[0117] Figure 4An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute an improved infrared small target detection method, which includes:

[0118] S1: Acquire and preprocess an image to obtain a preprocessed image;

[0119] S2: An infrared small target detection network model is established based on the YOLOv9 framework. The infrared small target detection network model includes a backbone network, a multi-scale feature fusion network, and a detection head module. The backbone network includes an SD_BS module, and the multi-scale feature fusion network includes a GSELAN module.

[0120] S3: Setting training parameters, training the infrared small target detection network model, and obtaining an infrared small target detection training model;

[0121] S4: Input the detection image to be detected into the infrared small target detection training model to obtain the detection result.

[0122] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0123] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0124] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0126] It should be noted that the embodiments of the present disclosure may be implemented by hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic: the software portion may be stored in a memory and executed by an appropriate instruction execution system such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-described apparatus and methods may be implemented using computer executable instructions and / or contained in a processor control code, such as a programmable memory or a data carrier such as an optical or electronic signal carrier providing such code.

[0127] In addition, although the operation of the method of the present disclosure is described in a particular order in the accompanying drawings, this does not require or imply that these operations must be performed in this particular order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flow chart can change the order of execution. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution. It should also be noted that the features and functions of two or more devices according to the present disclosure can be embodied in one device. Conversely, the features and functions of a device described above can be further divided into being embodied by multiple devices.

Claims

1. An improved infrared small target detection method, characterized in that: include: S1: Acquire and preprocess an image to obtain a preprocessed image; S2: An infrared small target detection network model is established based on the YOLOv9 framework. The infrared small target detection network model includes a backbone network, a multi-scale feature fusion network and a detection head module. The backbone network includes an SD_BS module, the multi-scale feature fusion network includes a GSELAN module, and the steps of the SD_BS module are: S11: For sizes Input image, input features ,Will After the transformation is split by step size in the size dimension, the slices are divided into a series of sub-features. The formula is: ...， ...， in, is the number of pixels of the length and width of the input image, is the number of channels, Indicates that the input features are in the long dimension with the Line starts to So far, in the wide dimension Column starts to So far, Extract pixels for the step size, according to The obtained sub-features are , ; S12: Splice the sub-features according to the dimension of the channel to obtain a spliced ​​feature ; S13: for the splicing features Perform BN batch normalization to obtain normalized splicing features; S14: Use SiLU activation function to perform nonlinearity on the normalized splicing feature to obtain output feature ,in ; S3: Setting training parameters, using the preprocessed images to train the infrared small target detection network model, and obtaining an infrared small target detection training model; S4: Input the detection image to be detected into the infrared small target detection training model to obtain the detection result.

2. The improved infrared small target detection method according to claim 1, characterized in that: The SD_BS module is used to replace the changeable kernel convolution layer in the backbone network.

3. The improved infrared small target detection method according to claim 1, characterized in that: The GSELAN module replaces the RepNCSPELAN4 module of the multi-scale feature fusion network.

4. The improved infrared small target detection method according to claim 1, characterized in that: The steps of the GSELAN module include: S21: Input module inputs data; S22: Convolve the module input data with the first module hyperparameter to obtain a first intermediate value; S22: performing channel splitting on the first intermediate value to obtain a second intermediate value; S23: input the second intermediate value and the second module hyperparameter into the GSCSP_1 module, and perform self-convolution to obtain a third intermediate value; S24 inputs the third intermediate value and the second module hyperparameter into the GSCSP_1 module, and performs self-convolution to obtain a fourth intermediate value; S25: performing channel splicing on the first intermediate value, the second intermediate value, the third intermediate value, and the fourth intermediate value to obtain a spliced ​​intermediate value; S26: Performing deconvolution on the concatenated intermediate value to obtain output data.

5. The improved infrared small target detection method according to claim 4, characterized in that: The steps of the GSCSP_1 module include: S201: Convolve the input data of the GSCSP_1 module to obtain the intermediate value of the first module; S202: Input the input data of the GSCSP_1 module into the GSBottleneck module to obtain the intermediate value of the second module; S203: Channel-joining the intermediate value of the first module and the intermediate value of the second module, and then performing self-convolution to obtain output data of the GSCSP_1 module.

6. The improved infrared small target detection method according to claim 5, characterized in that: The steps of the GSBottleneck module include: S2001: Convolve the input data of the GSBottleneck module to obtain the first convolution result; S2002: performing DSC on the first convolution result to obtain a second convolution result; S2003: concatenating the first convolution result and the second convolution result, and then performing shuffling processing to obtain a third convolution result; S2004: Add the first convolution result and the third convolution result to obtain the output result of the GSBottleneck module.

7. The improved infrared small target detection method according to claim 1, characterized in that: The inspection of the detection head module uses three scale feature inspections, including: a large scale inspection greater than or equal to 80 pixels; a medium scale inspection greater than or equal to 40 pixels and less than 80 pixels; and a small scale inspection greater than or equal to 20 pixels and less than 40 pixels.

8. An improved infrared small target detection system, used to execute an improved infrared small target detection method as claimed in any one of claims 1 to 7, characterized in that: include: Image preprocessing module: acquire and preprocess the image to obtain the preprocessed image; Network model building module: Based on the YOLOv9 framework, an infrared small target detection network model is established. The infrared small target detection network model includes a backbone network, a multi-scale feature fusion network, and a detection head module; the backbone network includes an SD_BS module, and the multi-scale feature fusion network includes a GSELAN module; Model training module: setting training parameters, using the preprocessed images to train the infrared small target detection network model, and obtaining an infrared small target detection training model; Image detection module: Input the detection image to be detected into the infrared small target detection training model to obtain the detection result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the improved infrared small target detection method as claimed in any one of claims 1 to 7 are implemented.

Images