BO-densenet: A bilinear one-dimensional densenet network based on multi-scale feature fusion for wood NIR classification

Abstract

With the rapid development of deep learning techniques, convolutional neural networks have been widely used in the field of spectroscopy. In this paper, a bilinear branching Densenet network model (BO-Densenet) based on multi-scale feature fusion is constructed by applying a one-dimensional convolutional neural network to classify six woods: Tung wood, Balsa wood, Poplar wood, PVA-modified Poplar wood, Nano-silica-sol modified Poplar wood, and PVA-Nano-silica-sol modified Poplar wood. The results show that BO-Densenet achieves 98.90% accuracy in classification on the test set, which is higher than 82.09% of Partial Least Squares, and also higher than 89.88% of Lenet, 93.56% of Alexnet, 94.12% of Resnet-18 and 96.69% of Densenet-40 when compared with other deep learning algorithms. This shows that the BO-Densenet proposed in this paper can accurately achieve wood classification and has potential application prospects.