In the urbanization of China, construction land is generally distributed as a continuous area, and its change shows distinct spatial aggregation leading to the strong spatial autocorrelation, which is more obvious in remote sensing imagery with a fine spatial resolution. Based on the TripleSat-2 multi-spectral remote sensing images covering Nanjing city in 2016 and 2017, the paper compares and analyzes the performance of remote sensing change detection of construction land after we introduced the spatial autocorrelation index of the change vector as the image feature. Firstly, the three typical spatial autocorrelation indices of local G, Moran's I and Geary's C are extracted, and then the optimal spatial range of Lag and the optimal autocorrelation index suitable for the change detection are determined. The results showed that: (1) The spectral change vector had significant positive correlation. (2) The optimal range of Lag can be determined by global Moran's I and semi-variance. (3) We used the local G and local Moran's I with the spectral change vectors to increase the F1 score of the change detection, and the local G showed the better performance. (4) Using the local G with the optimal range of Lag as additional image features, the F1 scores were 20% higher than that using only the spectral change vectors. Fusing local spatial autocorrelation information especially of the multi-scale local G as the additional image features can effectively improve the change detection accuracy of construction land in remfote sensing imagery.