Resolution-dependent shape classification of Jumunjin and Ottawa sand particles

This dataset contains categorical particle-shape classifications assigned to 7,874 sand particles at five image scales by two independent classification methods. Two materials were imaged by dynamic image analysis: angular Jumunjin standard sand (3,759 particles) and rounded Ottawa sand (4,115 particles). Full-resolution binary particle images were downsampled by factors of 0.2, 0.4, 0.6, 0.8 and 1.0, and each resulting image was classified into Krumbein–Sloss roundness (R1–R9) and sphericity (S3–S9) classes by a Fourier-descriptor method and by the computational-geometry method of Zheng and Hryciw (2015). Every particle contributes exactly ten records, one per method and scale, together with the particle-image size in pixels at that scale. The data support an analysis of the minimum particle-image size required for reliable shape classification, and show that the required size differs between descriptors and between particle morphologies. Empty label fields record measurement failures and are informative: their frequency rises sharply as image size decreases. Files are UTF-8 CSV. Column definitions, the image-naming scheme, acquisition-batch structure and known limitations are documented in README.md.