A rural construction land extraction algorithm for UAV images based on improved Gaussian mixture model and Markov random field

In this paper, we propose a novel rural construction land extraction algorithm for Unmanned Aerial Vehicle images using an improved Gaussian mixture model. Firstly, in the Gaussian mixture model, instead of mixed probability of various types of surface features, we calculate the prior probability of the various features in the neighborhood of each pixel using Markov random field. It can reflect the features´ spatial correlation. Secondly, we use the simulated annealing to obtain the global optimum parameter estimates in the process of parameter estimation. Finally, we calculate the posterior probability of each pixel for the features using the parameters´ estimated value. Then, we can obtain the spatial distribution of various features. The effect of the proposed algorithm is analyzed through experiment. The experiment shows that our proposed method can improve accuracy of construction land information extraction and has better performance than other methods.