Evaluation and Mitigation of Domain Shift Impact between Volumetric Submicro-Scale and Micro-Scale Computed Tomography Systems in the Context of Automated Binary Wood Classification

Rapid and reliable automated identification of wood species can be a boon for applications across wood scientific context including forestry and biodiversity conservation, as well as in an industrial context via requirements for timber trade regulations. However, robust machine learning classifiers must be properly analyzed and immunized against domain shift effects. These can degrade the automated system performance for input data variations occurring in many real-world scenarios. This work methodologically analyses the domain shift generated by using two differing sub-micro-scale and micro-scale computed tomography setups in the focused context of deep learning based binary wood classification from volumetric image data. To counteract this, we examine several mitigation strategies and propose primary data-level and narrow model-level strategies to effectively and successfully minimize the performance domain gap. Core elements of the data-wise strategy include the combined usage of phase-correction methods, low-pass pyramid representation of the data and adjustments of model normalization and regularization. Vanishing domain performance differences led to the conclusion that the combined strategy ultimately prompted the model to learn robust features. These features are discriminative for the utilized wood species data from both sub-micro-system and micro-system domains, despite the substantial differences in data acquisition setup that propagate into fundamental image quality metrics like resolution, contrast and signal-to-noise ratio.