Background Modeling Combined With Multiple Features in the Fourier Domain for Maritime Infrared Target Detection

Abstract

To detect ship targets in dynamic sea backgrounds, a robust foreground detection method based on background modeling combined with multiple features in the Fourier domain (BMMFF) is proposed. Because the fluctuation of seawater is similar to a sine wave, the amplitude spectrum of the seawater in the Fourier domain has strong energy concentration, and the background model is built in the Fourier domain. The local statistical characteristics can effectively reflect the properties of a specific position, and the target frequency points are extracted by comparing the local statistical characteristic differences between the amplitude spectrum of the test frame and the updated background. Additionally, two different strategies are used to update the background for cases with or without sudden waves in two adjacent test frames. When the target and seawater have different contrast in different scenes, the threshold is set according to the seawater's fluctuation degree, which is classified into three different levels in the training stage. Moreover, the linear correlation feature between the test frame and the updated background and the oscillation feature of the test frame's amplitude spectrum are proposed in the Fourier domain, which can better segment targets in relatively calm and violent sea scenes, respectively. Thus, the two features are combined with the background model with a reasonable strategy to further improve the detection accuracy. The experimental results demonstrate that BMMFF outperforms related comparison algorithms in different challenging sea scenes. The average false alarm rate of BMMFF is reduced by approximately 30% in the case where its average detection rate is similar to that of other algorithms.