FIST-Toy: A Benchmark Dataset for Toy Component Classification

摘要

As a comprehensive technology of machinery, electronics and computer vision, industrial automation technology has been attracting extensive attention and developed rapidly recently years. The refinement of target classification plays an important role in the industrial automation, therefore, many computer vision based target classification methods have been proposed. However, there is few real industrial dataset can be used to evaluating the target classification algorithms. In this paper, we construct a novel toys dataset, named FIST-Toy, including 7411 images of 31 categories. Base on the proposed dataset, we conduct a comprehensive study of the state-of-the-art pretraining target classification models qualitatively and quantitatively. In addition, we propose a weight voting based classification neural network. By comparing with AlexNET, VggNet, Inception and ResNet, the proposed model outperforms in fine-grading. We evaluate the FIST-Toy dataset using various deep learning models. And the experimental results show that the FIST-Toy dataset provides a benchmark to evaluate the performance of different methods for target classification.