Abstract: In the marine Internet of Things （M-IoT）， unmanned aerial vehicles （UAVs） offer flexible deployment and wide-area coverage， significantly enhancing the efficiency of data collection and edge computing. However， due to their limited computational and energy resources， UAVs struggle to handle high-density data processing demands independently. Meanwhile， traditional orthogonal multiple access techniques suffer from low spectral efficiency， making them insufficient for managing the high-concurrency data uploading demands of maritime sensing devices （SDs）. To address these challenges， this paper proposes a joint optimization strategy for UAV-assisted data collection and computation offloading based on non-orthogonal multiple access （NOMA）. In the proposed framework， the UAV serves as an aerial edge computing node and uses NOMA to simultaneously collect data from multiple maritime SDs. After SDs complete data transmission， the UAV performs part of the collected data locally while offloading the remaining to a ground base station， enabling air-ground collaborative computing. This approach alleviates local resource limitations and enhances the overall system processing capability. Based on this architecture， we formulate a UAV-assisted data collection and computation offloading problem for M-IoT， aiming to minimize the total system energy consumption by jointly optimizing SD and UAV transmission power， UAV flight trajectory， computational resource allocation， and the data offloading ratios. To solve the resulting non-convex optimization problem， we develop a hybrid solution framework that combines deep deterministic policy gradient （DDPG） with block coordinate descent （BCD）. DDPG is employed to optimize the UAV trajectory and transmission power strategies， while BCD handles the optimization of the offloading ratios and computational resource allocation. Simulation results demonstrate that the proposed joint optimization algorithm achieves significant energy efficiency improvements across various scenarios. Compared with baseline schemes， the proposed approach reduces total system energy consumption by up to 23.58%， effectively improving the energy efficiency of data collection and processing in complex marine environments.