 This paper proposes a novel 3D convolutional neural network, CNN, approach for automatic classification of crops from spatiotemporal remote sensing images. The 3D-CNN framework was designed to capture the full crop growth cycle, while also preserving the spatial and temporal information. An active learning strategy was introduced to select the most informative samples for training, which improved the accuracy of the model. Experiments showed that the 3D-CNN outperformed traditional 2D-CNNs, as well as other conventional methods. This article was authored by Shunping G, Chi Zhang, An Jinsu, and others.