Laser spot detection is a critical component of imaging laser warning systems, and its detection accuracy directly affects system response speed and platform survivability. Under complex backgrounds and low signal-to-noise ratio (SNR) conditions, conventional threshold- or edge-based detection methods often suffer from blurred boundaries, missed detections, and center drift. Although deep learning methods offer strong representation capabilities, they rely heavily on large volumes of annotated data and high computational resources. To address these issues, this study proposes a laser spot detection algorithm based on similarity measurement, which integrates physical modeling of laser transmission with image-domain spot matching, thereby achieving accurate spot detection and sub-pixel localization under complex backgrounds.

First, a quantitative mapping relationship among spot gray level, laser emission power, and transmission distance was established based on experimental data (Fig.3). On this basis, a dual-station imaging laser warning model was constructed to solve for the emission power and transmission distance of the incident laser (Fig.2). This model used the received powers measured by two spatially separated warning devices, together with the imaging response characteristics of the focal-plane array detector and the atmospheric transmission model, to obtain the emission power and transmission distance parameters. By further incorporating the above mapping relationship, a theoretical spot template corresponding to the actual scene (Fig.4) was generated (Fig.6). Finally, the theoretical template was used to perform sliding matching on the measured image. Coarse spot localization was achieved by calculating the normalized cross-correlation (NCC) between the theoretical template and the measured image (Fig.7), followed by refinement of the spot center using the gray-level centroid method, thereby improving detection robustness and localization accuracy under complex backgrounds and uneven illumination.

Experimental verification was conducted under typical clear-sky land conditions, with a visibility of approximately 15 km and a relative humidity of 30%~50%. The imaging system employed an indium gallium arsenide short-wave infrared focal plane array (INGAAS FPA) with a resolution of 640 pixel × 512 pixel, a pixel pitch of 15 μm × 15 μm, and a sampling bit depth of 16 bits. The two warning devices received optical powers of 213.67 W and 4.24 W, respectively. Based on the dual-station imaging laser warning model, the emission power and transmission distance of the incident laser were obtained, and the corresponding theoretical spot template was generated. Comparative results with the Nobuyuki Otsu method (OTSU), Canny edge detection (CANNY), and the watershed algorithm showed that, when the background was complex or the spot contrast was low, traditional methods tended to produce blurred boundaries, missed detections, or center deviation (Fig.5). In contrast, the physically modeled spot template combined with NCC matching produced a sharp and prominent correlation peak, significantly improving the discriminability of the spot location. After coarse localization, the gray-level centroid method was applied to further improve accuracy. Experiments on multiple spot images with complex backgrounds showed that the localization error of the proposed method was less than 1 pixel in all tests, whereas the three conventional algorithms exhibited larger deviations, noticeable drift, and even false detections (Fig.5, Table 4). Additional tests under different background textures, low SNR conditions and non-uniform illumination further verified the robustness of the method (Table 5). Overall, the proposed method demonstrated stable detection performance across a variety of complex environments, achieved sub-pixel accuracy, and outperformed conventional methods in both accuracy and robustness.

The proposed laser spot detection method based on similarity measurement effectively addresses the issue of insufficient detection accuracy in imaging laser warning systems under complex backgrounds and low SNR conditions. By integrating physical modeling with image matching, the method significantly improves the robustness and center localization accuracy of spot detection, and provides reliable theoretical and technical support for incident laser direction inversion and threat identification.