Maritime infrared small target detection plays a very important role in the field of safeguarding marine security and marine rescue. The long-distance ship target degenerates into a small spot in an infrared image, which has the characteristics of small size, weak intensity, serious lack of texture information, and is easily disturbed by noise. Meanwhile, the background clutters in infrared images presented by missile-borne infrared imaging system under backlighting conditions, such as sea-sky horizon line, island clutters, dense sea glints and sea surface bright band-like zone with high grayscale seriously affect the detection performance of long-distance infrared small targets, which can easily lead to target missed detection and false alarms. Only using one strategy to filter an infrared image may result very poor performance of small weak target detection. The reason for this situation is that the high-intensity clutter will interfere with the small target detection in the low-intensity clutter regions and the targets will interact with each other in the detection process. Therefore, using different strategies to filter different local regions of an infrared image is an effective target signal detection scheme. This paper mainly focuses on the key technologies that develop a high detection performance solution of infrared small targets under the complex maritime background. In this paper, the maritime scene perception information is constructed by sensing the regions where the non-stationary sea surface background clutters components exist and their fluctuation states, and the signal processing method adapted to the background clutter is adopted to refine the detection of small targets. Aiming at the characteristics of marine infrared imaging, the background clutters of the non-stationary maritime surface are perceived and the information of infrared marine scene is constructed. Firstly, the large eigenvalue of the structure tensor and Hough transform of an infrared image are used to detect the sea-sky horizon line. The connected region labeling algorithm operates the binary difference eigenvalues map to obtain the sea-sky island region. Bernaola-Galavan segmentation algorithm and the complexity represented by variance weighted information entropy are used to perceive the background clutter components such as sea glints and sea surface bright band-like zone. Therefore, the scene information of sea-sky island region, flat sea region, fluctuation sea region and island region is constructed. Secondly, aiming at the characteristics of clutter components, adaptive signal processing is used to suppress clutter and enhance the signal-to-clutter ratio of target. A clutter suppression and target enhancement algorithm based on directional difference is proposed to suppress the background and highlight target signal for sea-sky region and island region with strong edge structure. Top-hat transform is used to suppress clutter in flat sea area. In the fluctuation sea region of sea glints and sea surface bright band-like zone, a fusion gray gradient clutter suppression method is proposed to suppress the background, greatly reducing the interference caused by sea clutters. Finally, after clutter suppression, target extraction is carried out by targeted methods. The detection results of each region are normalized to the same intensity range and integrated. Constant false alarm threshold segmentation strategy is used to detect targets in sea-sky region, island region and fluctuating sea region. Whereas, the local peak decision strategy is used to detect the targets for flat sea region. Compared with the existing advanced infrared image small target signal detection methods, the experimental results show that the proposed single-frame detection method based on maritime scene perception can effectively adapt to different maritime background clutters, and can not only effectively increase the signal-to-clutter ratio gain and the background suppression factor, but also improve the accuracy and robustness of small target detection in complex backgrounds.