Green is the new Black: Multimodal Noisy Segmentation based fragmented burn scars identification in Amazon Rainforest

Detection of burn marks due to wildfires in inaccessible rain forests is important for various disaster management and ecological studies. The fragmented nature of arable landscapes and diversecropping patterns often thwart the precise mappingof burn scars. Recent advances in remote-sensingand availability of multimodal data offer a viablesolution to this mapping problem. However, thetask to segment burn marks is difficult because ofits indistinguishably with similar looking land patterns, severe fragmented nature of burn marks andpartially labelled noisy datasets.In this work we present AmazonNET – a convolutional based network that allows extracting of burnpatters from multimodal remote sensing images.The network consists of UNet- a well-known encoder decoder type of architecture with skip connections. The proposed framework utilises stackedRGB-NIR channels to segment burn scars from thepastures by training on a new weakly labelled noisydataset from Amazonia.Our model illustrates superior performance by correctly identifying partially labelled burn scars andrejecting incorrectly labelled samples, demonstrating our approach as one of the first to effectivelyutilise deep learning based segmentation models inmultimodal burn scar identification.

Back to AI for Social Good event