Light absorption and scattering in turbid underwater environments can substantially degrade image quality, thereby limiting the accuracy of fish target detection, individual tracking, and behavioral parameter extraction. To meet the demand for automated observation of fish behavior in aquaculture waters, this study focuses on underwater videos of Decapterus maruadsi observed beneath aquaculture rafts in Dongshan Bay. Based on underwater polarization imaging and a physical underwater degradation model, we developed an integrated technical workflow encompassing image restoration, video enhancement, target detection, trajectory tracking, and behavior analysis. First, indoor polarized image datasets and field underwater video datasets were used to analyze image degradation characteristics under varying turbidity levels, observation distances, and complex illumination conditions. An image restoration method for turbid waters was then established to improve image clarity, texture details, and overall visual distinguishability. On this basis, the static image restoration strategy was extended to continuous video processing, incorporating exposure control and inter-frame consistency constraints to form a video restoration workflow suitable for aquaculture raft scenarios. Subsequently, a YOLO-pose model was employed to simultaneously obtain target bounding boxes and body keypoints of Decapterus maruadsi, while the ByteTrack multi-object tracking algorithm was used to achieve continuous association of individual trajectories. Furthermore, based on trajectory and keypoint information, behavioral parameters including speed, acceleration, turning angle, tail-beat amplitude, vertical movement, inter-individual distance, and group spatial distribution were extracted, enabling quantitative characterization of individual movement patterns and group behavioral features in Decapterus maruadsi. This study effectively links underwater image restoration with fish behavior analysis and provides an extensible methodological framework for automated fish behavior monitoring and underwater observation in turbid aquaculture waters.