ACTL: Asymmetric Convolutional Transfer Learning for Tree Species Identification Based on Deep Neural Network

The identification of tree species is great significance to the sustainable management and utilization forest ecosystems. Hyperspectral data provide sufficient spectral spatial information classify species. Convolutional neural networks (CNN) have achieved success in hyperspectral image (HSI) classification. outstanding performance CNN HSI classification relies on training samples. However, it’s expensive time consuming acquire labeled In this article, a novel asymmetric convolutional transfer learning model for proposed. First, dataset built from Goddard’s LiDAR, & Thermal (G-LiHT) data. Then, weights trained ImageNet are used initialize model. Finally, well fine-tuned network perform task. experimental results reveal that proposed with blocks effectively improves accuracy Howland provides new idea remote sensing images.