<p>Infrared ship target detection is a fundamental task in computer vision that is of great significance for ship monitoring and marine safety. Ship detection based on infrared polarization images has the advantages of strong anti-interference capabilities and operating in all-weather conditions compared to detection based on visible and traditional infrared images. However, the existing deep learning-based infrared polarization ship target detection algorithms suffer from inadequate usage of polarization characteristics and misdetections caused by complex scenarios. This paper introduces a forward and backward propagated polarization feature extraction module based infrared ship target detector (ISTD). Firstly, a polarization feature enhancement module based on an improved polarization bidirectional reflectance distribution formulation was built in ISTD to enable more robust polarization feature extraction, which can improve the interpretability of the following forward and backward propagated neural networks. Then, a novel forward propagation structure for feature up-sampling and down-sampling processes was proposed, which deepens the feature channels while taking polarization angle variables into account for polarization degree map formulation. Finally, a backward propagation method with an embedded polarization optimization scheme was introduced, which uses the polarization degree map error as feedback to regulate the learning process. Besides, error bound analysis under consecutive failures is provided for ISTD. This paper also presents an infrared polarization ship target dataset (IPSTD) comprising over 9,000 images. Extensive experiments conducted using the IPSTD evaluate the favorable performance of ISTD and indicate that it achieves excellent detection results across various challenging infrared ship detection scenarios when compared to nine state-of-the-art (SOTA) detectors.</p>