Part 5/12:

He highlighted the shift towards FPGA-based processing—semiconductors more suitable for resource-constrained environments—though deploying complex ML models on FPGAs remains an evolving challenge due to their current limitations in deep learning compatibility.

Edge AI Model Development and Optimization

Akash revealed that building and deploying ML models on satellites involves multiple steps: pre-processing images, tiling large multispectral images, cloud detection, and segmentation. Because satellite images can reach hundreds of megapixels with multiple spectral bands, tiling is essential to manage processing complexity.