Individual Tree Species Classification Based on a Hierarchical Convolutional Neural Network and Multitemporal Google Earth Images

Accurate and efficient individual tree species (ITS) classification is the basis of fine forest resource management. It a challenge to classify in dense forests using remote sensing imagery. In order solve this problem, new ITS method was proposed study, which hierarchical convolutional neural network (H-CNN) model multi-temporal high-resolution Google Earth images were employed. an experiment conducted park Beijing, China, GE several significant phenological phases broad-leaved forests, namely, before after mushrooming period, growth wilting selected, classifications based on these along with typical CNN models H-CNN conducted. experiment, accuracy multitemporal higher by 7.08–12.09% than those single-temporal images, offered OA 2.66–3.72% models, demonstrating that rich features species, together model, can effectively improve classification.