Low Earth Orbit (LEO) satellite networks offer a promising solution for providing global coverage and efficient communication services to terrestrial users. However, due to the highly dynamic and time-varying nature of the constellation topology, together with frequent inter-satellite link switching, the direct application of conventional terrestrial network routing algorithms in LEO satellite networks may result in suboptimal performance since they cannot quickly adapt to the dynamic satellite network topology and varying traffic loads. In this paper, we propose an inter-satellite transmission scheme based on Raptor Codes (IST-RC), leveraging the rateless and low-overhead properties of Raptor codes to mitigate packet loss and out-of-order delivery caused by satellite handover, buffer overflow, and time-varying link conditions in LEO satellite networks. Considering the extended propagation delay inherent to inter-satellite links, we propose an inter-satellite transmission scheme based on systematic Raptor codes (IST-SRC) to mitigate resource under-utilization observed in the IST-RC scheme. Furthermore, we propose a lightweight reinforcement learning-based environment-adaptive algorithm to dynamically adjust key parameters of the IST-RC and IST-SRC schemes. Theoretical models are derived to analyze the delay performance of the proposed approaches, and numerical results confirm both the accuracy of these models and the efficiency of the lightweight RL-based transmission schemes in LEO satellite network scenarios.