The Terrestrial Ecosystem Carbon Inventory Satellite (TECIS/CM-1) utilizes a combination of multi-beam lidar, multi-spectral cameras, and other passive and active sensors for synergistic observations, enabling high-resolution, comprehensive, and three-dimensional atmospheric monitoring of clouds and aerosols. In recent years, traditional algorithms have faced challenges in terms of vertical layer retrieval accuracy and robustness in complex environments with low signal-to-noise ratios, near-surface observations, and mixed multi-layer structures. To address these issues, this paper proposes TECIS-CASNet, a generalized framework for atmospheric layer recognition and application, designed for the novel multi-beam lidar on the TECIS satellite, leveraging the characteristics of the lidar data and deep learning attention mechanisms. To validate the reliability of this framework, the research team conducted multiple ground-based synchronous observation experiments to systematically evaluate its recognition accuracy. Finally, as a demonstrative application, the study focuses on a typical long-distance dust transport event in the Beijing-Tianjin-Hebei region of China, showcasing the practical application value of the framework. The results indicate that the TECIS-CASNet framework achieves high cloud-aerosol recognition accuracy, reaching 98.41%, and is capable of reducing misidentification and missed detection in complex environments, including low signal-to-noise ratios, near-surface layers, and multi-layer mixed structures. The absolute accuracy of aerosol optical depth retrieval is 0.01, with an overall accuracy of 98%. This paper, centered around the TECIS-CASNet framework, provides significant insights for lidar satellite atmospheric remote sensing data processing and environmental monitoring applications.