WPCNet: Highly Accurate Workpiece Counting Network in Unrestricted Scenes

The development of steel/aluminum (called work-pieces) industry often focuses on productivity, labour costs and quality control, which are essential for industry. However, the counting of workpieces is inseparable from a large amount of labor, so that the problems of low efficiency, high cost and low accuracy of manual counting are highlighted. To address these issues, we propose an accurate workpiece counting network base on deep learning, named WPCNet, which aims to achieve fully automatic workpieces counting tasks in unrestricted scenarios. In response, we collected 10 classes workpiece images as a dataset, and perform a series of preprocessing for apply it to the algorithm model used in this paper. In other words, the number of workpieces in the final image is the number of contours output by the model. Extensive experiments on the newly created workpiece dataset demon-strate the effectiveness of WPCNet, improving social work efficiency.