Exploring the Feasibility of Hailo 8L for Embedded Multispectral Classification

Multispectral remote sensing image classification typically imposes high computational demands, limiting its use for real-time applications on resource-constrained platforms. This work presents an embedded AI framework for efficient remote sensing image classification using a Raspberry Pi 5 equipped with a Hailo-8L AI chip accelerator. Two deep learning architectures, a lightweight CNN and a more complex GAN-based ResBaGAN model, were adapted and optimized for execution in this framework, achieving significant inference speedups while maintaining competitive accuracy. Experiments conducted on multispectral images demonstrate that Hailo-8L enables the Raspberry Pi to achieve execution times comparable to those of a more powerful desktop device. The results confirm the feasibility of deploying sophisticated remote sensing classification models on low-power embedded systems, opening new possibilities for on-site, cost-effective environmental monitoring applications.

Palabras clave: Raspberry Pi, Hardware accelerator, Remote Sensing, Classification, Deep Learning, Embedded computing