In this paper we address the image denoising problem specifically for high resolution multispectral images. We explore the possibility to device a method that approximates the denoising function using the ideology of learning from data. The overall objective is to build a model that replicates the denoised results of Non-local means algorithm using far less computational resources for each of the four spectral bands (blue, green, red and near infrared). We have used deep neural networks and in particular stacked autoencoders to learn the function that maps a noisy image to its denoised version. We show that after training the model on a large set noisy and denoised images, excellent results that are comparable to the results of Non-local means algorithm can be obtained in much less computational time. The scope of the model can further be extended to denoise any other natural image by training it on the appropriate data set.