Table 4.

Overall performances (%) of the proposed PND-Net built upon different standard base CNNs.

Dataset	Base CNN + GCN	Top-1 accuracy	Top-3 accuracy	Precision	Recall	F1-score
Banana	ResNet-50	90.00	98.34	90.00	90.00	90.00
	Xception	89.25	98.27	90.00	89.00	89.00
	Inception-V3	83.77	98.13	84.00	84.00	84.00
	MobileNet-V2	83.99	97.80	84.00	84.00	83.00
Coffee	ResNet-50	89.52	97.00	89.00	89.00	89.00
	Xception	90.54	98.67	90.00	90.00	90.00
	Inception-V3	89.18	98.67	89.00	89.00	89.00
	MobileNet-V2	89.86	98.33	90.00	89.00	89.00
Potato	ResNet-50	94.32	99.03	94.00	94.00	94.00
	Xception	96.18	99.42	96.00	96.00	96.00
	Inception-V3	96.05	99.64	96.00	96.00	96.00
	MobileNet-V2	92.59	98.68	93.00	93.00	93.00
PlantDoc	ResNet-50	84.11	98.02	85.00	84.00	84.00
	Xception	84.30	98.10	85.00	84.00	84.00
	Inception-V3	81.00	98.05	81.00	81.00	81.00
	MobileNet-V2	80.81	97.86	81.00	81.00	81.00
BreakHis 40$\times$	ResNet-50	95.50	99.00	95.00	95.00	95.00
	Xception	94.83	99.00	95.00	95.00	95.00
	Inception-V3	95.00	99.00	95.00	95.00	95.00
	MobileNet-V2	94.00	99.00	94.00	94.00	94.00
BreakHis 100$\times$	ResNet-50	96.79	99.00	97.00	97.00	97.00
	Xception	95.19	99.00	95.00	94.00	94.00
	Inception-V3	95.67	99.00	96.00	96.00	96.00
	MobileNet-V2	95.83	99.00	96.00	96.00	96.00
SIPaKMeD	ResNet-50	99.18	100.00	99.00	99.00	99.00
	Xception	98.98	100.00	99.00	99.00	99.00
	Inception-V3	98.37	100.00	98.00	98.00	98.00
	MobileNet-V2	98.17	100.00	98.00	98.00	98.00

The best top-1 accuracy (%) achieved on each dataset is given in bold.