The first Terrestrial Ecosystem Carbon Monitoring (TECM-1) satellite—Gou Mang was successfully launched on August 4, 2022 in China. TECM-1 satellite is equipped with a multi-beam LiDAR which is mainly used to obtain the elevation of land and forest. The mission of measuring the elevation of a forest requires both high-ranging accuracy and horizontal positioning accuracy. The multi-beams LiDAR on the TECM-1 satellite is concurrently equipped with an optical axis monitoring camera to obtain the footprint images, including the ground images and spot images. The centroid of the laser spot in the spot image can indicate the laser pointing angle, helping reduce the horizontal positioning errors. However, the laser spot is inevitably influenced by the background features in the spot image, and traditional centroid extraction methods are difficult to achieve ideal accuracy. Many scholars have made improvements based on traditional methods, and most of them apply noise reduction processing to the spot image, such as grayscale threshold constraint and image filtering. However, due to the influence of photography conditions and complex terrain, it is challenging for these relatively simple noise reduction methods to separate the laser spot from background noise. The centroid extraction accuracy still faces challenges in complex background scenarios. We report a laser spot centroid extraction method against complex background noise, which can effectively remove background noise from the spot image. It achieves better adaptability and higher accuracy than previous methods. We hope that our research can provide a certain reference value for the future processing and application of laser data collected by the TECM-1 satellite.

A set of footprint images collected by the TECM-1 satellite consists of the ground image and spot image and the spot image can be considered to be formed by the laser spot and surrounding background features. The ground image and spot image are geometrically aligned and exhibit noticeable brightness differences due to varying exposure time (Fig. 2). Based on the characteristics of footprint images, the present study firstly crops spot area image pairs and non-spot area image pairs from footprint images. Then, the grayscale transformation coefficients are calculated by non-spot area image pairs, and the background grayscale values can be removed from the spot image. Subsequently, a distance constraint is adopted to limit the range of laser spots and a Gaussian filter is applied to smooth the spot image. Then an adaptive threshold is estimated by Otsu's method for binary segmentation. It generates a binary mask of the laser spot and with the mask processing, the noise of the spot image can be almost eliminated. Finally, the centroid coordinates of the laser spot are calculated by the grayscale centroid method. The proposed method is tested with simulated and actual spot images, and the accuracy of centroid extraction is analyzed along with the stability of the laser spot centroids.

For the generated 10215 simulated laser spots, our method exhibits significantly improved centroid extraction accuracy compared to the Gaussian fitting method, grayscale centroid method, and ellipse fitting method. It results in an average error of approximately 0.059 pixel, and a root mean square error of about 0.074 pixel, slightly higher than those of Ren's method. The maximum error is 0.482 pixel, better than Ren's 1.828 pixel (Table 2). Furthermore, the CE90 of the proposed method is approximately 0.11 pixel (Fig. 6). In addition, our method is approximately 60% faster than Ren's method and comparable to the ellipse fitting method in processing time (Table 3). When applied to 1000 sets of consecutive footprint images obtained within 25 s, our method suggests that the standard deviation of centroid coordinates is less than 0.05 pixel. The range is within 0.2 pixel in a single direction (Table 5), demonstrating strong stability (Fig. 7 and Fig. 8). We select over 40 scenes of footprint images at intervals of 5 d to analyze the stability of centroids, and the results show that the standard deviations of centroid coordinates distribute between 0.12 and 0.34 pixel. It corresponds to horizontal distances of about 1 to 2.7 m on the ground, signifying that the relative geometric relationship between the laser and optical axis monitoring camera is relatively stable (Table 6). Additionally, the stability of centroid coordinates varies from different laser beams in the monitoring range (Figs. 9-13). It can be found that regular monitoring for the stability of centroids is necessary to comprehend centroid variations.

We report a laser spot centroid extraction method against complex background noise based on the footprint images collected by the TECM-1 satellite. Our method uses grayscale matching to remove most of the background noise from the spot image, followed by distance constraint, Gaussian filtering, and Otsu's image segmentation to remove the residual noise. The method ultimately calculates the centroid coordinates of the laser spot by the grayscale centroid method. Assessed by simulated and actual spot images of various terrain types, the results demonstrate that our method exhibits robust adaptability to laser spots in complex background scenarios and displays strong stability. However, for ground images with a significant number of saturated grayscale values, the present grayscale matching method cannot remove the background noise completely and it will affect the final centroid extraction accuracy. Future research should explore optimizing the grayscale transformation model or introducing other constraints to address this issue. The results of the stability analysis reveal a relatively tight jitter scope of centroids during the monitoring range, indicating a relatively stable geometric relationship between the laser and optical axis monitoring camera. Our study will be employed to support laser positioning, with the goal of enhancing the horizontal positioning precision of laser data.