CN-121982285-A - Infrared small target detection device and method based on Mamba state space model

Abstract

The application discloses an infrared small target detection device and method based on Mamba state space model, and relates to the field of infrared image processing. The infrared detection system comprises a data preprocessing module, a model construction module, a model training module and a result output module, wherein the data preprocessing module is used for preprocessing an original infrared image to obtain an infrared image set, dividing the infrared image set into a training set, a verification set and a test set, the model construction module is used for constructing an infrared small target detection model based on a Mamba-state space model, the model training module is used for inputting the infrared images of the training set and the verification set into the infrared small target detection model to perform target detection training, and the result output module is used for inputting the infrared image of the test set into the trained infrared small target detection model to obtain a detection result of the infrared small target. The method is used for realizing efficient and accurate infrared small target detection under limited computing resources.

Inventors

• ZHOU DONGJIE

Assignees

• 北京遥感设备研究所

Dates

Publication Date

20260505

Application Date

20251230

Claims (10)

1. 1. An infrared small target detection device based on Mamba state space model, which is characterized by comprising: The data preprocessing module is used for preprocessing an original infrared image to obtain an infrared image set, and dividing the infrared image set into a training set, a verification set and a test set; the model construction module is used for constructing an infrared small target detection model based on Mamba state space models; the model training module is used for inputting the infrared images of the training set and the verification set into the infrared small target detection model to perform target detection training; and the result output module is used for inputting the infrared image of the test set into the trained infrared small target detection model to obtain the detection result of the infrared small target.
2. 2. The method of claim 1, wherein the model building module comprises a feature extraction network, wherein: the feature extraction network is composed of a plurality of serially connected selective state space modules, and the selective state space modules are used for carrying out long-sequence feature modeling through a linear attention mechanism and state space reconstruction.
3. 3. The method of claim 2, wherein the selective state space module comprises a selective state space block and a local feature enhancement unit, wherein: the selective state space block comprises two parallel processing paths, wherein one path comprises linear transformation and SiLU activation functions, and the other path comprises linear transformation, state space reconstruction, selective gating and layer normalization; The local characteristic enhancement unit adopts a residual structure and is used for enhancing local information representation through multi-scale characteristic extraction.
4. 4. The method of claim 1, wherein the model building module comprises a multi-scale feature aggregation module, wherein: the multi-scale feature aggregation module comprises a state space reconstruction unit and a selective gating mechanism unit, wherein: the state space reconstruction unit is used for calculating channel weights by using adaptive linear transformation based on state space reconstruction; The selective gating mechanism unit is used for generating a spatial feature weight map through a selective gating mechanism.
5. 5. The method of claim 4, wherein the multi-scale feature aggregation module further comprises a feature fusion unit: The feature fusion unit is used for carrying out feature fusion by adopting element-by-element multiplication.
6. 6. The method of claim 1, wherein the apparatus further comprises a model optimization module: The model optimization module is used for optimizing the infrared small target detection model by adopting a self-adaptive loss function and a state space regularization strategy.
7. 7. The method of claim 1, wherein the model building module comprises a detection head: The detection head comprises a classification branch, a regression branch and a centrality branch, wherein: The classification branch is used for predicting the target class probability; the regression branch is used for predicting the target boundary box offset; The centrality branches are used for predicting target centrality scores.
8. 8. The method of claim 7, wherein the detection head is of a lightweight design.
9. 9. A method based on the infrared small object detection device of claims 1-8, comprising: preprocessing an original infrared image to obtain an infrared image set, and dividing the infrared image set into a training set, a verification set and a test set; constructing an infrared small target detection model based on Mamba state space models; Inputting the infrared images of the training set and the verification set into the infrared small target detection model to perform target detection training; And inputting the infrared image of the test set into a trained infrared small target detection model to obtain a detection result of the infrared small target.
10. 10. An electronic device, comprising: A processor, and a memory arranged to store computer executable instructions that, when executed, cause the processor to perform the steps of the method of claim 9.

Description

Infrared small target detection device and method based on Mamba state space model Technical Field The application relates to the technical field of infrared image processing, in particular to an infrared small target detection device and method based on Mamba state space model. Background The infrared small target detection has important application value in the fields of military reconnaissance, security monitoring and the like. Traditional infrared small target detection methods rely mainly on Convolutional Neural Networks (CNNs) and transducer architectures. CNNs have a good local feature extraction capability, but have limitations in capturing long-distance dependencies. Although a transducer can establish global correlations, the computational complexity is high and distraction problems easily occur when dealing with small objects. In recent years, a State Space Model (SSMs) represented by Mamba exhibits excellent performance. Compared with the traditional method, SSMs has linear computational complexity, and is more suitable for being used as the basis of a lightweight model. SSMs has unique advantages in modeling sequence data and capturing long-range dependencies, which is of great significance for detecting small targets in infrared images. In practical applications, the infrared detection device usually needs to consider the limitation of computing resources, so that developing a lightweight and efficient detection model has important practical value. Disclosure of Invention The application aims to provide an infrared small target detection device and method based on Mamba state space model, which are used for realizing efficient and accurate infrared small target detection under limited computing resources. In order to achieve the above purpose, the application adopts the following technical scheme: In one aspect, the present application provides an infrared small target detection device based on Mamba state space model, including: The data preprocessing module is used for preprocessing an original infrared image to obtain an infrared image set, and dividing the infrared image set into a training set, a verification set and a test set; the model construction module is used for constructing an infrared small target detection model based on Mamba state space models; the model training module is used for inputting the infrared images of the training set and the verification set into the infrared small target detection model to perform target detection training; and the result output module is used for inputting the infrared image of the test set into the trained infrared small target detection model to obtain the detection result of the infrared small target. Preferably, the model building module comprises a feature extraction network, wherein: the feature extraction network is composed of a plurality of serially connected selective state space modules, and the selective state space modules are used for carrying out long-sequence feature modeling through a linear attention mechanism and state space reconstruction. Preferably, the selective state space module includes a selective state space block and a local feature enhancement unit, wherein: the selective state space block comprises two parallel processing paths, wherein one path comprises linear transformation and SiLU activation functions, and the other path comprises linear transformation, state space reconstruction, selective gating and layer normalization; The local characteristic enhancement unit adopts a residual structure and is used for enhancing local information representation through multi-scale characteristic extraction. Preferably, the model building module comprises a multi-scale feature aggregation module, wherein: The multi-scale feature aggregation module comprises a state space reconstruction unit and a selective gating mechanism unit: the state space reconstruction unit is used for calculating channel weights by using adaptive linear transformation based on state space reconstruction; The selective gating mechanism unit is used for generating a spatial feature weight map through a selective gating mechanism. Preferably, the multi-scale feature aggregation module further comprises a feature fusion unit: The feature fusion unit is used for carrying out feature fusion by adopting element-by-element multiplication. Preferably, the apparatus further comprises a model optimization module: The model optimization module is used for optimizing the infrared small target detection model by adopting a self-adaptive loss function and a state space regularization strategy. Preferably, the model building module includes a detection head: The detection head comprises a classification branch, a regression branch and a centrality branch, wherein: The classification branch is used for predicting the target class probability; the regression branch is used for predicting the target boundary box offset; The centrality branches are used for predictin