Improvement Strategies for Mask R-CNN in Satellite Image Analysis

With the development of mega-constellations and space technology, the frequency and number of satellite anomalies in orbit are increasing in tandem, and the demand for autonomous in-orbit repair operations is becoming increasingly urgent. Autonomous repair technology is limited by the efficiency and accuracy of identifying and operating on repair targets (such as malfunctioning parts and capture components) in orbit, and it is not yet capable of supporting in-orbit application implementation. To address this issue, this paper proposes a satellite component recognition method based on an improved MASK R-CNN model, which introduces a new network structure, CB-Resnet, and integrates the Particle Swarm Optimization (PSO) algorithm to achieve precise fault part recognition and localization. The CB-Resnet module enhances the model's ability to express key component features in images, while the PSO algorithm optimizes network parameters, improving the model's global search capability and generalization performance. With this method, repair spacecraft can quickly identify operational components, thereby supporting the formulation of rapid fault handling plans, improving repair efficiency, and reducing operational difficulty. Experimental results show that the improved model performs excellently in component recognition tasks, providing an effective technical support for the development of autonomous in-orbit satellite repair operations.