Finding Fire¶

Client: William Makant, Plumis Ltd, william@plumis.com.

Plumis develops fire-suppression systems using 32×24 thermal images to detect fires and determine where suppression should be directed. The current approach relies mainly on classification, using separate neural networks to estimate fire probability and fire location. This project will investigate whether spatial machine-learning techniques, such as object detection or segmentation, could combine both tasks in a single model. Students will develop an annotation workflow, train and evaluate suitable neural-network approaches, and assess their performance on very low-resolution thermal data. A key challenge is determining whether these methods can remain accurate while being sufficiently small and fast for deployment on resource-constrained embedded hardware.