AUTOMATED DETERMINATION OF MORPHOMETRIC INDICATORS OF BAST FIBER RAW MATERIALS FROM CROSS-SECTION MICROIMAGES

Abstract

Bast fiber raw materials are used in the textile, food, construction, and automotive industries, increasing the requirements for objective quality control of this material. Current assessment methods rely on separate physical procedures for each indicator and do not provide a unified spatial picture of fiber bundle distribution. The aim of the study is to improve the method for assessing bast fiber quality by developing an instance segmentation method for cross-section microimages, enabling objective calculation of morphometric indicators for determining fiber functional suitability. A dataset was formed from 83 microscopic cross-sections with 1984 expertly annotated fiber bundle instances. An instance segmentation model was developed and trained on the YOLO11s-seg architecture with a singleclass detector and 768-pixel input resolution. Domain-specific augmentations and morphometric post-processing calculated the area and centroid of each detected bundle. Three morphometric indicators were proposed: equivalent fiber diameter, structural fiber fraction, and degree of defibration. The model achieves mask mAP@0.5 of 0.900 on the validation set, with balanced precision (0.842) and recall (0.833). Analysis time per microimage is approximately 33 ms, independent of bundle count. The proposed method calculates all three indicators from a single microimage, whereas standardized methods such as ASTM D7025-09 and ASTM D6961-09 require separate physical procedures for each indicator. Automated segmentation eliminates manual bundle counting and provides measurement accuracy and reproducibility unattainable through manual analysis with an eyepiece micrometer. The results confirm the method's suitability for objectively determining the functional suitability of bast fiber raw materials for food and textile industry applications, including the production of filtering materials.