A transfer learning-based convolutional neural network and its novel application in ship spare-parts classification

Spare parts play an essential role in the logistical support of heavy-asset industries that require highly reliable and efficient operations. This paper focuses on spare-parts classification using transfer learning theory via deep convolutional neural networks. Through the present literature, multi-criteria spare-parts classification is innovatively realized using image-processing methods with a three-phase model. Neural network architecture was reconstructed based on the pre-trained VGG-19 in the first step. And the hierarchical inference model of spare parts was subsequently built with a purpose of data visualization and image conversion. In the final phase, the proposed model was trained and tested using actual data generated from a real-world case which demonstrated its excellent performance with an average accuracy of 96.36% in the five-fold cross-validation. To further verify the model, a comparative scheme was proposed using different convolutional neural network-based learning methods, including well-known AlexNet and ResNet-50. The proposed model's performance is outstanding with an overall accuracy of 95.87%. This novel transfer learning-based application for the multi-criteria classification problem successfully achieves an adjustable solution framework for spare-parts management.

Automated summary

The study classified spare parts using deep convolutional neural networks and transfer learning theory, achieving an average accuracy of 96.36% through a three-phase model. The proposed model demonstrated superb performance with an overall accuracy of 95.87% compared to other neural network learning methods.