LAUSR

Since the geometric characteristics of curved workpieces are prone to reflection, the severe reflection phenomenon will adversely affect the task of defect detection on the workpieces surface. In order to solve the reflective problem of curved workpieces, a multiexposure fusion method based on deep learning is proposed in this article. This method is an unsupervised end-to-end information metric densely connected network (IMDCN). At the same time, a novel dataset of curved workpieces for the multiexposure fusion task, called the CW-MEF dataset, is established. The IMDCN method combines spatial attention and channel attention mechanisms with the backbone network DenseNet. The fusion results generated directly by the network are fed into the loss function for joint training along with the source images and information metric weights. To demonstrate the performance of the method in detail, four different types of image fusion metrics are used as evaluation standards for fusion results. The qualitative and quantitative experimental results with seven state-of-the-art MEF methods show that our method exhibits satisfactory performance and time efficiency.