Estimating boreal forest ground cover vegetation composition from nadir photographs using deep convolutional neural networks

Hilary A. Cameron, Pranoy Panda, Martin Barczyk, Jennifer L. Beverly

Abstract: HIghlights•Forest ground cover and surface vegetation were documented with smartphone photos.•The downward (nadir) photos were manually segmented into 10 discrete cover types.•Percent cover from manually classified pixels correlated well with field measurements.•A deep convolution neural network (DCNN) was trained to segment cover automatically.•The DCNN segmentation had 95% accuracy and independent validation showed promise.