Abstract: This paper analyzed the probability distribution of pixel gray of the following 5 kinds of objects such as building, farmland, garden plot, woodland and water. The results showed that the probability distributsion of pixel gray of the 5 kinds of objects was not standard normal distribution, but approximately normal distribution. This paper revised the training samples with Gaussian normal function, the probability distribution of pixel gray of the revised training sample was standard normal distribution. So condition probability density function of class was modified, the classification accuracy for maximum-likelihood algorithm increased by 5.25%.