Cloud detection is an important and difficult task in the processing of satellite remote sensing data. Clouds are one of the most major noise presents in the satellite image. Presence of clouds hide the ground information. So, it is not possible to predict the values of pixels affected by clouds. Satellite images are having wide variety of applications. Because of the presence of clouds, the satellite images are partially unusable. To use the information from satellite images it is necessary to detect the presence of clouds. Accurate detection of clouds is particularly important before using satellite images for applications like land use classification, object detection, forest fire detection, climate change analysis, etc. In this paper a novel method for automatic cloud detection based on U-Net convolutional neural network is proposed. Qualitative and quantitative evaluations are conducted on publicly available cloud validation datasets, which indicates that our proposed method is more effective and accurate than the state-of-the-art methods and can accurately detect clouds under different conditions.