Using Deep Convolutional Networks for Species Identification of Xylotheque Samples

Forest species identification is critical to scientifically support many environmental, commercial, forensic, archaeological, and paleontological actividades. Therefore, it is very important to develop fast and accurate identification systems. We present a deep CNN for automated forest species identification based on macroscopic images of wood cuts. We first implement and study a modified version of the LeNet convolutional network, which is trained from scratch with a database of macroscopic images of 41 forest species of the Brazilian flora. With this network we achieve a top-1 accuracy of 93.6%. Additionally, we fine-tune the Resnet50 model with pre-trained weights on Imagenet to reach a top-1 accuracy of 98.03%, which improves previous published results of research on the same image database.