To improve the resilience of distribution networks (DNs) in the event of extreme natural disasters such as typhoons and rainstorms, it is imperative to efficiently implement distribution service restoration (DSR) to restore loads as soon as possible. In previous studies, DSR has mainly adopted the distributed resource model with droop or PQ control. This inhibits the exploitation of the potential of distributed generators (DGs) in load restoration when the DN loses support from the upstream transmission network. Thus, this paper proposes a multi-resource collaborative service restoration (MRCSR) approach for DNs incorporating local soft open points, DGs, and tie switches. The MRCSR model is developed by integrating a decentralized hierarchical droop control (DHDC) strategy and incorporating the frequency and voltage features of the load demand. A two-stage iterative feedback optimization (TSIFO) algorithm is then developed to analyze the MRCSR model in an accurate and efficient manner. Finally, the proposed model and algorithm are tested on the modified IEEE 33-bus system and a practical distribution system of the Taiwan Power Company to verify their effectiveness and advantages over existing approaches.