Methane Segmentation on FPGA

This project focuses on the onboard optimization of machine learning pipelines for methane segmentation on FPGA edge devices. The primary goal is to address and solve aerospace data-downlink bottlenecks by processing data directly at the edge.

The core pillars of the research include comparing Post-Training Quantization (PTQ) versus Quantization-Aware Training (QAT). The impact of INT8 quantization is extensively evaluated in terms of memory footprint, processing throughput, and power consumption on the ZCU104 FPGA utilizing Vitis AI.

Finally, the implemented solutions are benchmarked against low-power CPUs using the MARS-EMIT dataset to demonstrate the effectiveness and efficiency of the FPGA-based approach.