Abstract: In recent years, polarimetric imaging technology has rapidly advanced and demonstrated unique advantages in applications such as target detection, material identification, and surface property analysis. However, conventional linear polarimetric imaging techniques can only capture partial polarization information, and their theoretical models struggle to effectively distinguish target characteristics under weak polarization signals in low-altitude scenarios. This paper proposes a novel target detection algorithm, YOLOv8-FPZ, based on full-Stokes polarimetric images. By integrating an enhanced MPFNet network that incorporates circular polarization information, the algorithm employs dual-directional CGabor convolutional kernels designed in this work to comprehensively capture polarization features of light fields, thereby significantly enhancing image details and contrast. A dedicated full-Stokes polarimetric dataset is constructed to support this research. The FPZHead detection head integrates improved Distributed Feature Learning and a reparameterized convolutional structure, enabling effective extraction of both local details and global features from polarimetric images and enhancing the capability to process challenging samples. Additionally, the C2F module is replaced with a DBB)convolution, combining multi-scale branches to enrich single-convolution representations, while the loss function is optimized to Focaler-GIoU for enhanced robustness. Experimental results demonstrate that YOLOv8-FPZ outperforms the baseline YOLOv8, achieving improvements of 7.1% and 3.2% in mAP@0.5 for aircraft and pedestrian detection, respectively. This algorithm addresses the limitations of single linear polarimetric imaging and exhibits high detection efficiency and accuracy for small targets in low-altitude scenarios.