AgrUNet: A Multi-GPU UNet Based Model for Crops Classification

Agriculture acts as a catalyst for comprehensive economic growth, boosting income levels, mitigating poverty, and contrasting hunger. For these reasons, it is important to monitor agricultural practices and the use of parcels carefully and automatically to support the development of sustainable use of natural resources. The deployment of high-resolution satellite missions, like LandSat and Copernicus Sentinel, combined with AI Deep Learning (DL) methodologies has revolutionized Earth Observation science, enabling studies on yield predictions, soil classifications, and crop mappings on large areas, and the analysis and processing of Big Data using innovative approaches. This approach requires high-performance computing systems since DL algorithms are known to be very computing-heavy, and recent multi-GPU HPC systems can boost by one or two orders of magnitude the processing power of classical computing systems based only on CPUs. In this study, we develop AgrUNet, a scalable, fast, and reliable UNet-based architecture DL model to perform crop classification on multispectral multitemporal satellite data, implemented and optimized to run on single and multi-GPU HPC systems. Our model achieves a Dice score of approximately 0.90, a peak throughput of 59 and 605 /s for the train and inference steps respectively, improving by approximately a factor 7X the best results reported in the literature and quite ideal speedup running both on a 4X V100 and 8X A100 GPU systems.