The primary aim of this project is to deploy, adjust, and advance fire smoke detection algorithms and models that were developed and tested in previous projects (P2.38 and P2.61). These models will be utilized in an on-orbit demonstration on the Kanyini and Phi-Sat-2 missions, with the goal of validating their accuracy, speed and power consumption in specific scenarios. Additionally, the project seeks to investigate domain adaptation strategies to adapt a model initially trained on simulated images. This method will employ a limited number of Kanyini or Phi-Sat-2 images overactive fires to begin the adaptation process, with an incremental learning approach used to progressively enhance the model’s accuracy as more real fire smoke images are collected.

Moreover, advancements of the existing AI Model and On-Orbit Processing will be explored to improve detection accuracy and performance. The project also aims to develop live demonstrations to showcase advancements and engage key stakeholders, thereby supporting the commercialization pathway and ensuring broad collaboration.