CFAR Algorithm Based on Different Probability Models for Ocean Target Detection

Abstract

The two-parameter constant false alarm rate (CFAR) detection algorithm uses the background average $u_{b}$ and the standard deviation $\sigma _{b}$ to determine the target detection threshold, which is simple and easy to implement. However, it is limited by the assumption of Gaussian distribution. Based on different probability distributions, the research improves the two-parameter CFAR algorithm, and proposes the two-parameter CFAR method based on initial detection, and the detection methods based on Loglogistic Distribution model and Adjoint Covariance Correction Model (ACCM). Three methods are used to detect and extract ocean targets in the same research area, and the results are compared and analyzed. The experimental results show that ACCM proposed in the research fits the long tail characteristic of the ocean background under complex ocean conditions well. Its goodness of fit is improved by nearly 50% compared with Loglogistic Distribution model, and its amount of false alarm of ocean target detection is 77.78% of Loglogistic model. In addition, in view of a large amount of calculation caused by the sliding window of the traditional two-parameter CFAR, OceanTDA9 deep learning model is designed for initial detection in research, which improves the detection speed of ocean targets.

Keywords

• Ocean target detection
• two-parameter CFAR
• loglogistic distribution
• deep learning model
• initial detection