Quantitative wood anatomy (QWA) is a field of wood anatomy that characterizes the variability of xylem anatomical features in trees, shrubs, and herbaceous species. QWA analyzes functions, growth, and environment related to wood anatomical features by...
Quantitative wood anatomy (QWA) is a field of wood anatomy that characterizes the variability of xylem anatomical features in trees, shrubs, and herbaceous species. QWA analyzes functions, growth, and environment related to wood anatomical features by measuring their dimensions, numbers, and distribution. However, there has been difficulties to produce large data sets of xylem anatomical data. Although image acquisition and analysis techniques have been improved in their performance and easy-of-use functions, methods of quantification of the anatomical features are still not feasible to produce large amount of data, which is a basis of statistical representativeness.
In this presentation, we explored capability of an instance segmentation technique as a promising quantification method to obtain massive amount of data. To elucidate merits of the instance segmentation technique over threshold-based image analysis, both methods were applied for detecting and segmenting pores and tyloses in microscopic images of Quercus aliena Bl.
Threshold-based image segmentation and analysis is ready to use for separating pore-like objects in an image, but not capable of distinguishing anatomical features such as pore and tylosis without further analysis. Also, requirement of manual parameter optimization makes automatic segmentation and measurement difficult for large number of images. On the other hand, the instance segmentation technique requires elaborate preparation of training datasets and optimization to detect and segment most of pores and tyloses, but, after optimization, the trained model is versatile for detection and segmentation of the anatomical features. In addition, the technique is robust against various illumination and sample conditions. Utilizing the instance segmentation technique, we can automate collection of quantitative data from massive microscopic images of anatomical features of wood. Therefore, the instance segmentation technique can be a powerful tool for quantitative wood anatomy. - 44