Automated banknote inspection systems commonly rely on optical or multispectral sensing for denomination recognition, counterfeit screening, and note fitness assessment. This paper investigates whether a compact 77-81 GHz frequency-modulated continuous-wave (FMCW) millimeter-wave (mmWave) radar can detect damaged notes inside a bundled stack and further characterize damage burden and damage configuration. Using a Texas Instruments AWR1843BOOST radar front-end with DCA1000EVM raw-data capture, measurements were collected from bundled U.S. currency under four viewing geometries: Bottom-Horizontal, Bottom-Vertical, Top-Horizontal, and Top-Vertical. The dataset was organized into three machine-learning tasks: (1) binary damage detection, (2) damage-burden classification, and (3) damage-configuration classification. Two feature granularities were evaluated: capture-level descriptors derived from full background-subtracted range profiles and window-level descriptors derived from local profile segments. Across pooled geometries, capture-level damage detection achieved a balanced accuracy of 0.9821 and a macro F1-score of 0.9709, while window-level damage detection reached 1.0000 for both metrics. Damage-burden estimation improved from a capture-level balanced accuracy of 0.7708 to a window-level balanced accuracy of 1.0000 when scalar features were augmented with delta-profile information. Exact damage-configuration recognition remained the most difficult task, with pooled balanced accuracies of 0.4062 at the capture level and 0.4375 at the window level, yet the confusion matrices showed structured sensitivity rather than random failure. These results indicate that mmWave FMCW radar is highly promising for non-contact banknote damage screening and burden-aware inspection, while still retaining partial sensitivity to configuration-specific damage patterns.
Onta Shahriar (Wed,) studied this question.