Mapping Forest Burn Extent from Hyperspatial Imagery Using Machine Learning

Support vector machines are shown to be highly effective in mapping burn extent from hyperspatial imagery grasslands. Unfortunately, this pixel-based method is hampered forested environments that have experienced low-intensity fires because unburned tree crowns obstruct the view of surface vegetation. This obstruction causes misclassified as unburned. To account for misclassifying areas under crowns, trees surrounded by can assumed been burned underneath. effort used a mask region-based convolutional neural network (MR-CNN) and support machine (SVM) determine pixels post-fire forest. The output classifications MR-CNN SVM were identify image vegetation pixels. These also label being within fire’s extent. approach results higher accuracy eliminating false negatives burns obscured achieving nine percentage point increase accuracy.