A spatial point target recognition algorithm combining image registration and improved K-means clustering

There are few spatial point target features in deep space background images. To achieve accurate target recognition, a spatial point target recognition algorithm is proposed by utilizing the motion difference between stars and targets, and combining image registration and improved K-Means clustering algorithm. The algorithm first determines the differences between adjacent images caused by motion, eliminates the motion of stars between frames through image registration, and then uses the IR-Kmeans clustering algorithm to classify stars, noise, and targets, completing the recognition of spatial point targets. Finally, the performance of the point target recognition algorithm was validated by simulating distant spatial images. The experimental results show that the line of sight angle corresponding to the recognized target has an error of less than 0.07 ° from the true value, and the frame rate processed by the algorithm is 10 fps. Compared with existing clustering based spatial point target recognition algorithms, the proposed algorithm does not require adjustment of the relevant parameters of the algorithm for in orbit missions, and can compensate for the shortcomings of existing clustering algorithms that cannot be applied to high orbit scenarios, enhancing the autonomy and applicability of the algorithm.