Show Auto-Adaptive and Tell: Learned From the SEM Image Challenge

• Main Authors: Jing Su, Jing Li
• Format: Article
• Language: English
• Published: IEEE 2021-01-01
• Series: IEEE Access
• Subjects: Scanning electron microscopy; show; adaptive and tell model; adversarial training
• Online Access: https://ieeexplore.ieee.org/document/9383261/

Scanning electron microscopy (SEM) has been widely used in optical material science. However, a considerable quantity of human resources is required to analyze and describe SEM images. In recent years, the application of computer technology in material science and engineering developed endlessly. Computer science, including data processing, simulation technique, and mathematical model, promotes material science progress tremendously. Moreover, deep learning has been achieved success in image classification and image analysis. In this paper, we propose a novel automatic analysis tool using a triplet neural network called show auto-adaptive and tell to analyze optical SEM images automatically. Firstly, we collected SEM images and corresponding captioning from previous papers and built a database. Then, a triplet neural network with proposed loss function to train the show auto-adaptive and tell model on 60% of the dataset for SEM images analysis, test on 30% and validate on 10%. Finally, experiment on the four metrics index as the evaluation criterion shows that the novel method gets better performance than previous work.