Following technology advances in high-performance computation systems and fast growth of data acquisition, a technical breakthroughnamed Deep Learning made remarkable success in many research areas and applications. Nevertheless, the progress of hardwaredevelopment still falls far behind the upscaling of deep neural network (DNN) models at the software level. NGA seeks to apply neuralnetwork miniaturization techniques on geospatial data. To address this critical need and limitations of state-of-the-art approaches, IntelligentAutomation, Inc. (IAI), along with our collaborators, proposes to develop several novel model-compressing techniques applicable for neuralnetworks for geospatial data. The key innovations of the proposed approach include Structured Sparsity Learning (SSL) to regularize thememory and data pattern, omnipresent error back propagation (Omni-BP) technique targeted for activation sparsification, and QuantizedGradients (QGrad) to improve the efficiency of the distributed learning. To overcome the issue of training data availability, we will leverage IAIsHumanView (HumV) software to generate synthetic data on scenarios interesting to National Geospatial Intelligence Agency (NGA) for thetraining purpose.