Multi-label tree species classifier for satellite imagery. See ForestClassifier.pdf for detailed description.
TLDR:
Input - satellite image of forested area.
Output - list of tree species contained in the image.
Architecture - CNN and ResNet-50.
Data - Sentinel-2 images of Maine, USA; United States Forest Service Forest Inventory and Analysis.
Results - A binary accuracy of 83% was achieved.
Examples:
The input image can be seen. The true species in the image and the predicted species can be seen underlined in red.
Maine:
Alabama (completely distinct dataset):
Serbia (no true data available):
