傅里叶域内双背景模型的海上红外序列图像目标检测

针对不同波动程度的海场景下红外目标检测的难点，提出一种傅里叶域内对海背景进行双模型建模的目标检测算法。由于不同的海场景具有不同的海水波动模式，用单一的模型较难准确的描述所有的海场景。海水在傅里叶域内幅度谱的稳定性较强，因此提出一种傅里叶域内的双海水背景模型，即概率单高斯模型和混合高斯模型。首先对海上红外图像进行行方向的傅里叶变换，获得图像的幅度谱和相位谱。然后利用训练阶段纯海水图像幅度谱序列的方差对海水的波动程度进行判别。若海水波动较剧烈，利用傅里叶域内的概率单高斯模型对海水进行建模。若海水波动较平缓，利用傅里叶域内的混合高斯模型对海水进行建模。将测试帧幅度谱和海背景模型进行比较，获得检测结果。实验结果表明，本算法在不同类型的海场景下，均能有效的抑制海背景和准确的检测出目标。 

Target Detection for Maritime Infrared Sequence Image Based on Dual Background Modeling in the Fourier Domain

Aiming at the difficulty of infrared target detection in sea scenes with different fluctuation degree， an algorithm based on dual modeling for sea background in the Fourier domain is proposed. Different sea scenes have different fluctuation degree， and it is difficult to accurately describe all the sea scenes with a single model. The stability of the seawater’s amplitude spectrum in the Fourier domain is strong. Therefore， a dual seawater background model in the Fourier domain， namely mixture Gaussian model and probabilistic single Gaussian model， is proposed. Firstly， Fourier transformation in the row direction is performed for maritime infrared images to obtain the amplitude spectrum and phase spectrum. Next， the variance of the amplitude spectrum sequence of pure seawater images is used to identify the fluctuation degree of seawater in the training stage. If the seawater fluctuates violently， the probabilistic single Gaussian model in the Fourier domain is used to model the seawater. If the seawater fluctuates gently， the mixture Gaussian model in the Fourier domain is used to model the seawater. By comparing the amplitude spectrum of the test frame with the sea background model， the detection results are obtained. The experimental results show that the algorithm can effectively suppress the sea background and accurately detect the target in different sea scenes.