FiberNet: auto-detection of multi-fibers in microscopic image using instance segmentation

Abstract

The identification of animal fibers has always been an important topic in the textile field. Therefore, it is very important to identify fiber types accurately and quickly for textile fiber quality inspection. Although some traditional image-based fiber recognition techniques can identify one to four fibers in an image, the methods usually need to segment the fibers in the image into a single fiber image, which is a time-consuming process. Therefore, the FiberNet model is proposed to automate this task. Different from other methods, this study used the case segmentation model for the first time to detect multiple fibers and their species. FiberNet was developed based on BlendMask, introducing a denoising module and an attention module to improve the final detection accuracy and speed. Therefore, the model performance and identification accuracy of FiberNet are higher than those of the fiber detection algorithm, with identification accuracy of up to 99.51%.

Keywords:

• Image identification
• multi-fibers
• instance segmentation
• automatic detection

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China, under Grant 61876106, the Error! Hyperlink reference not valid., under Grant 18ZR1416600, the Zhihong Scholars Plan of Shanghai University of Engineering Science, under Grant 2018RC032017, and the Shanghai Local capacity-building projects, under Grant 19030501200.