Deep learning methods achieve very high classification accuracies in many tasks, including satellite image classification. However, these methods lack the transparency and simplicity of other classification algorithms. Sparse coding has emerged as an effective tool in classifying images, and provides the user with an efficient algorithm that easily relates the classification output to the original input feature space. In this work, we explore the viability and the effectiveness of a popular sparse coding algorithm, label-consistent k-means singular value decomposition (LC-KSVD), in classifying images from the satellite data set Sat-4. This paper provides a framework for using feature extraction, sparse coding, dictionary learning, and classifier training on the Sat-4 dataset, achieving a 94.5 % accuracy.