The work aims to summarize the packaging waste recovery routing as a reverse logistics vehicle routing problem with back path and time window (RL-VRPBTW), so as to construct a model with the minimum recovery cost, departure cost and time window penalty as the joint optimization objective. A better initial solution was obtained by introducing the factor of "vehicle remaining space recovery ability" to improve the classical heuristic C-W algorithm. Based on the Scatter Search framework, a Scatter Search algorithm (ISISS) based on initial solution improvement was designed. According to the problem model, the algorithm functions were realized through five steps of diversity generation method, reference set update method, subset generation, merging method and solution improvement method. In the city-type geographical scenario of "dense distribution of some recovery points", three scale examples with 50, 100 and 200 recovery nodes were randomly generated, and two vehicle types were considered for simulation experiments. The ISISS algorithm was compared with C-W, GA and SS algorithms to verify that the algorithm proposed had better performance in solving the routing problem of large-scale packaging waste recovery vehicles. The simulation results indicate that ISISS is a better algorithm to solve the multi-objective large-scale packaging waste recovery routing problem.