An automatic defect detection method for TO56 semiconductor laser using deep convolutional neural network. (May 2023)

• Record Type: Journal Article
• Authors: Zhang, Hang; Li, Rong; Zou, Dexiang; Liu, Jian; Chen, Ning
• Is Part Of: Computers & industrial engineering. Volume 179(2023)
• Publication Date: 2023-05
• Subjects: Deep convolutional neural network -- Defect detection -- Image segmentation -- Semiconductor laser -- TO56; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2023.109148
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A three-stage defect detection method is proposed for semiconductor laser TO56. An optimized DeepLab-v3+ is proposed for the wire image segmentation. An equipment is designed to automatically collect images and detect defects. Abstract: Accurate and reliable automatic defect detection of semiconductor laser plays an important role in the application and development of optical communication technology, but the traditional methods cannot meet the increasing requirements of defect detection of semiconductor lasers. In this study, an automatic three-stage defect detection method is proposed for the most widely used semiconductor laser TO56. The key defect detection tasks of TO56 include the sintering status detection of LD and PD and the connection reliability detection of four gold wires. In the stage I of the proposed method, the captured TO56 image is judged whether the chips and wires have been sintered by the object detection algorithm, and the wire images are extracted. Then in the stage II, the extracted wire images are segmented by the image segmentation algorithm. Finally, in the stage III, the category pattern recognition is performed based on the segmentation results. Moreover, to further improve the effectiveness of the proposed method, the object detection algorithm adopted in the stage I and the image segmentation algorithm adopted in the stage II are optimized. The experimental results demonstrate that the proposed method can accurately and automatically detect the defects of TO56, and the accuracy reaches 98.85 %.