Edge Detection of Microstructure Images of Magnetic Multilayer Materials via a Richer Convolutional Features Network

Magnetic multilayer materials are extensively used in micro-devices and nanoelectronics areas. It is significant to implement edge detection and extraction for the microstructure images of the multilayer materials. This research deals with the edge detection and extraction of microstructures images of the magnetic multilayer material based on a richer convolutional features (RCF) network. First, an RCF network model on a 20-fold expanded Berkeley Segmentation Data Set and benchmark 500 (BSDS500) dataset is retrained. Then, such model is applied to the edge detection test on the given microstructure images of the magnetic multilayer material, and the edge probability maps containing coarse and obvious boundaries between the layers of magnetic materials are obtained. Third, the non-maximum suppression (NMS) algorithm is introduced to further refine the thick edges of the microstructure images. The results demonstrate that the RCF-based edge detection method is capable of detecting light and unclear boundaries of the magnetic multilayer material from their images, and outperforms the existing other edge detection algorithms includes Canny operator and HED network. In addition, under the expanded RCF model combining with the NMS algorithm, the edge probability map of the microstructure images of the magnetic multilayer material are almost the same as the ground truth.