Application of neural networks for classifying softwood species using near infrared spectroscopy
Sang-Yun Yang, Ohkyung Kwon, Yonggun Park, Hyunwoo Chung, Hyunbin Kim, Se‐Yeong Park, In-Gyu Choi, Hwanmyeong Yeo
Abstract
Lumber species identification is an important issue for the wood industry. In this study, three types of neural networks (artificial neural network (ANN), deep neural network (DNN), and convolutional neural network (CNN)) were employed for classifying softwood lumber species using NIR spectroscopy. The results show that CNN, which is based on deep learning, was more stable than the other neural networks. In particular, the stability of the training process was remarkably improved in CNN models. During the training procedure, the validation accuracy of the CNN model was 99.3% for the raw spectra, 99.9% for the standard normal variate (SNV) spectra and 100.0% for the Savitzky-Golay second derivative spectra. Interestingly, there was little difference in the validation accuracies among the CNN models depending on mathematical preprocessing. The results showed that CNN is sufficiently adequate to classify the softwood lumber species.