Deep learning-based automated morphology classification of electrospun ultrafine fibers from M44 element image of muller matrix

Abstract

Electrospun ultrafine fibers with microporous morphology have been wildly used in the fields of drug release, filtering material, and tissue engineering, and it is more difficult to obtain the information of their microporous morphology. In this paper, electrospun ultrafine fibers with different morphologies: smooth surface, microporous, and beaded microspheres were prepared. Then, the polarized information of various fibers was obtained with polarized light microscope and calculate the Muller matrix. Subsequently, the method of transfer learning was introduced, to train the discriminant model with only a small amount of data based on M44 image element of the Mueller matrix, and to realize the automatic classification of the microporous morphology of electrospun ultrafine fibers. These results show that this classification method with a high accuracy in the test set could provide a fast, simple, reliable and real-time analysis for researchers to screen fiber samples with different morphologies.