RetinaNet-Based Compact Polarization SAR Ship Detection
Sheng Gao, Hongli Liu
Abstract
In this article, we propose a target detection algorithm with a polarization feature-driven neural network for compact polarimetric (CP) SAR data. First, we use the power–entropy (PE) decomposition theory to obtain the polarization features <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$h_{\mathrm {esa}}$ </tex-math></inline-formula> of the CP SAR data, then input <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$ h_{\mathrm {esa}}$ </tex-math></inline-formula> into the neural network RetinaNet for training, and finally perform prediction on the test dataset. We compare the proposed method in this article with the constant false alarm rate (CFAR) algorithm, and the experimental results show that the neural network-based target detection algorithm has fewer false alarms while ensuring detection accuracy.