Figure 1.

Classification accuracy of the hybrid high-order FC networks by integrating DN_L, DN_H and DN_A (DN_L + DN_H + DN_A) with different kernel weighting factors τ ₁ and τ ₂ for DN_L and DN_H, respectively (the weight factors for DN_A are determined by ${\tau }_3=1-({\tau }_1+{\tau }_2)$). The best accuracy was achieved with ${\tau }_1=0.3$ and ${\tau }_2=0.5$, which is 93.2% as indicated by an arrow.