Abstract text (incl. figure legends and references)

Crystalline polymers, such as polyethylene, are used in various applications because of their lightweight, low cost and excellent processability. The physical properties of crystalline polymers largely depend on their hierarchical solid-state structures with sizes spanning the nanometer to micrometer range. Understanding the hierarchical structure is essential for the structural design of crystalline polymeric materials. For this reason, it is desirable to have a method to analyze each level of the hierarchical structure at various scales.

Polarized optical microscopy (POM) has often been used to observe spherulites because the molecular orientation in the spherulite can be observed from the change in polarization state when linearly polarized light is incident on the sample. However, the resolution of POM is limited to a feature size of ∼250nm, making it difficult to detect small spherulites on a submicron scale. The limitation of POM can be complemented using electron microscopy, such as transmission electron microscopy (TEM). One of the technical problems associated with TEM observations is the ultra-thin sectioned samples, which are often deformed during microtoming.

We observed poly(ʟ-lactic acid) spherulite using POM and TEM in the same field of view. We attempted to combine the advantages of each technique - namely, molecular orientation information by POM and detailed scale morphology information by TEM [1]. To match the field of view of the TEM and POM images, a large number of TEM images were stitched together to produce a high-resolution large-area TEM image. After that, the deformations of the TEM ultrathin sections were corrected by reference to the POM images at the exact position of the sample, and large-area TEM images without deformations were successfully produced (Fig. 1).

We demonstrated a method to detect spherulites from the acquired large-area TEM image using YOLO, a well-known object detection system using convolutional neural networks. We obtained spherulites' number density and space-filling factor (relative crystallinity) within the desired region (Fig. 2). Such local structural information is key to understanding the nucleation and growth mechanisms of polymer spherulites. This detection process is speedy and much more powerful than manual processing and therefore is expected to analyze large amounts of image data. Furthermore, the present method would be more effective than the conventional methods in studying the heterogeneously distributed spherulites in materials (e.g., preferential nucleation at the filler surface in polymer composites).

References:

[1] Konyuba, Yuji, et al. "Correlative light and electron microscopy of poly (ʟ-lactic acid) spherulites for fast morphological measurements using a convolutional neural network." Microscopy 71(2) (2022): 104-110.

Fig. 1. Same field of view of PLLA by polarized microscope (POM) and transmission electron microscope (TEM). (a) POM image. (b) Large-area TEM image of the same field of view as the POM image (deformations caused during ultrathin section preparation have been corrected). (c) Merged image of the POM and TEM images.

Fig. 2. Results of spherulite detection by the convolutional neural network. (a) Image of a 250 µm × 250 µm area from the large-area TEM image. (b) TEM image after spherulite detection. Each red frame shows a detected spherulite and its width and height.