Abstract:

Objective: In order to solve the problems of strong subjectivity and low efficiency that the identification of furniture wood species mainly depends on manual identification in daily life, a common furniture wood species identification model based on Mobilenetv3 convolutional neural network (CNN) was designed to effectively improve the identification speed and accuracy of wood species. It provided a scientific and effective method for the rational utilization of wood resources, the management of wood import and export trade and the determination of furniture wood types by consumers. Method: Firstly, 3 880 images of four kinds of wood images coated without wood wax oil and two kinds of wood images coated with wood wax oil were collected. The data set was divided into training set, verification set and test set according to 6∶2∶2 and 2∶6∶2, and the data of the training set was expanded to four times of the original by using operations such as rotation and flip. Then four convolution neural networks and two traditional machine learning methods were used to establish the recognition model for the wood image without coated wood wax oil. Through analysis and comparison, the optimal recognition network model Mobilenetv3 was obtained, and the parameters of the model were optimized based on transfer learning. The wood species identification model based on MobileNetv3 network was constructed by putting the wood images coated without wood wax oil together with the wood images coated with wood wax oil and forming a new data set together with the remaining two kinds of wood images coated without wood wax oil. Finally, in order to simplify the operation of the classification personnel and reduce the operation difficulty of the staff in the actual detection, we selected the above wood species recognition model and built a wood species recognition system based on PyQt5. Result: Compared with traditional machine learning methods, convolutional neural network had better recognition results, and transfer learning could significantly improve the convergence speed and classification performance of the network. In the validation set, MobileNetV3 with the best recognition performance had an average recognition accuracy of 98.13% for four wood images without coated wood wax oil and 97.25% for wood images mixed with coated wood wax oil. Conclusion: In this paper, the wood species recognition model not only recognized the recognition of wood images coated without wood wax oil, but also realized the fast and accurate recognition of the painted wood wax wood image. Apart from bring convenience to wood classifiers, the wood species recognition model also could provide a reliable identification method for consumers to select solid wood furniture and determine wood species.

Key words: furniture wood, wax oil, identification, convolutional neural network, MobileNetV3