Abstract:To address computational inefficiency, high real-time processing demands, and elevated data security risks arising from the massive growth of water conservancy remote sensing data, this study proposed a multi-level secure computation offloading system based on cloud-edge-device collaboration. A joint optimization model encompassing offloading decisions, transmission power improvement, and computational resource allocation was constructed. A data security classification mechanism translated the security requirements of water conservancy data at different confidentiality levels into hard offloading constraints, aiming to minimize the weighted total cost of system energy consumption and latency while respecting maximum tolerable delay. Given that this joint optimization model was an NP-hard problem, an improved genetic algorithm was introduced, incorporating adaptive dynamic adjustment of crossover and mutation probabilities and a local search mechanism to enhance global optimization capability. Simulation results indicated that the improved genetic algorithm effectively overcame the tendency of traditional algorithms to be trapped in local optima, strictly ensured security constraints for classified data, and significantly reduced computational costs while improving processing efficiency. The proposed system achieves an optimal trade-off between performance and security, providing efficient and secure technical support for water management applications such as flood monitoring and water resource surveys.