Robotic Mastication Peanut Image Dataset for Shape-Descriptor-Based Structural Analysis

This dataset accompanies the paper "Shape-Descriptor-Based Structural Indices for Quantifying Food Breakdown in Robotic Mastication" by Ren et al., submitted to Advanced Engineering Informatics (2026).

This is version 2 of the dataset and supersedes version 1 (DOI: 10.5281/zenodo.19836500). It reflects the full set of analyses reported in the current manuscript.

The dataset contains 165 robotic-chewing peanut images (11 samples × 15 chewing cycles), the corresponding peanut-foreground RGBA crops, the per-image structural descriptor values used in the paper (PD, Solidity, LPI, ED, ENN_MN, AREA_CV), and Python scripts that reproduce all statistical analyses and figures reported in the paper. The analysis pipeline covers mixed-effects modelling, MANOVA, Pearson correlation, per-cycle summaries, principal component analysis (PCA), descriptor robustness diagnostics, smoothed-data inference, per-sample trend decomposition, two-stage trend inference, and per-sample slope analysis.

The dataset supports two reproduction paths: (1) reproducing the paper's statistical results and figures directly from the included CSV using the provided analysis scripts (no model weights required), or (2) reproducing the SAM and descriptor stages from the included foreground crops. See README.md for full instructions.

Note: The trained Mask R-CNN ResNet-101-FPN weights used for foreground extraction were not preserved and are therefore not included; users wanting to reproduce the foreground extraction stage will need to retrain the model. See models/README.md for details.