The application of community detection (community discovery) has been widely used in various fields for several years. To improve the algorithm accuracy, we proposed a memetic algorithm based on an adaptive simulated annealing local search (MA-ASA). Segmented label propagation (STLP) is used for initialization and variation operations. A hierarchical idea is adopted to form an initial cluster center during initialization, and random competition is used to select the next generation of solutions. Instead of using fixed probabilities in each crossover and variation operation, we used quality differences to switch to adaptive probabilities in simulated annealing (SA) for local search to accelerate convergence. The algorithm was extensively tested and experimented with 11 artificial and 4 real networks. Compared with other 10 algorithms, the results showed that MA-ASA performs well and is highly competitive.