To deal with the incompleteness of sensor network data, an attribute discretization method is proposed, based on the improved K-means clustering algorithm. In the method, a cluster index is defined and the data characteristics of each attribute are utilized in order to reduce the influences of noise, outliers and incomplete data sets on recognition results. Then, through the introduction of the mutual information entropy and variable precision correction coefficient, an attribute reduction heuristic algorithm is proposed, which integrates the advantages of the variable precision rough set and neighborhood rough set. The computational complexity is reduced and the recognition accuracy is improved by using the algorithm. Simulation results show the effectiveness of the proposed algorithms in dealing with the recognition accuracy and the computational complexity of the decision recognition system. The simulated environment test further verifies the applicability of the proposed algorithms.