Against the problem of a great distance between estimated distance and actual Euclidean distance caused by irregular network and network hole that eventually results in insufficient localization accuracy, an improved multidimensional scaling localization algorithm based on matrix correction and chaotic particle swarm optimization(CMDS-CPSO) is proposed. Distance among each pair of nodes is calculated by recursive strategy and further weighted by the received signal strength, so as to reduce the distance error between estimated distance and actual Euclidean distance as well as avoid the problem of network hole. Then chaotic particle swarm optimization is adopted to solve the parameter problem during the coordinate conversion process, which could loosen the influence of parameters to a high degree. Compared with the SPSO-MDS algorithm and MDS-DMC algorithm, the simulations reveal that the proposed algorithm of CMDS-CPSO could not only significantly improve the localization accuracy of nodes but has better robustness and adaptability to irregular networks.