A SAR image segmentation method based on salient feature constraints and level set

Abstract

Synthetic Aperture Radar (SAR) images suffer from issues such as speckle noise, intensity inhomogeneity, blurred edges, and complex interferences, making accurate SAR image detection a challenge. To address these challenges, this paper introduces a Saliency Feature-Constrained Level Set Segmentation Model (SFC-LSM). First, a Dual-Channel Pulse Coupled Neural Network detection model that integrates SAR image saliency features (SFF-DPCNN) is constructed. Using the SFF-DPCNN detection method, target features are initially extracted to capture the shape details. Subsequently, this shape prior information serves as the shape constraint term in the fine segmentation level set, thereby improving its energy function. The proposed method provides improved segmentation results for SAR images under complex interferences and exhibits robustness in handling background noise images. Comparative experiments are performed on SAR datasets with varying interference. The experimental results indicate that the proposed method is highly robust against speckle noise and interference, allowing for precise segmentation of ship targets.