fig4

Figure 4. Enhanced data quality directly gives rise to clear separability in the feature space. The t-SNE visualization (A) reveals that progressively improved data quality leads to increasingly clear class boundaries among different gesture groups; The monotonic decrease in DBI (B) further supports this trend. However, its correlation with recognition accuracy is relatively weak; In contrast, the FSI (C) exhibits a stronger correlation with accuracy, enabling more effective quantification of inherent separability (D). t-SNE: t-Distributed stochastic neighbour embedding; DBI: Davies–Bouldin Index; FSI: fundamental separability index; TSR: temporal sampling rate; SSR: spatial sampling rate.