Table 6 Performance assessment of proposed models based on the three datasets.

Model name	Model type	Complexity	Parameters	Memory	Inference time	Top-1 ACC	Top-5 ACC	Training time	Robustness	Cucumber		Banana		Tomato
Model name	Model type	(GFLOPs)	(M)	(M)	(s)	Top-1 ACC	Top-5 ACC	(hrs)	(Noise/Light)	ACC	FScore	ACC	FScore	ACC	FScore
AlexNet	CNN	0.61	54.16	674.2	1.25	0.860	0.950	20	Moderate	0.909	0.866	0.839	0.834	0.932	0.931
mAlexNet	CNN	0.73	68.04	819.3	1.56	0.845	0.930	22	Moderate	0.872	0.872	0.834	0.836	0.912	0.916
EfficientNetV2	CNN	8.40	22.10	76.80	4.41	0.870	0.940	48	High	0.897	0.888	0.757	0.754	0.935	0.935
MobileNetV3	CNN	0.52	5.40	25.60	3.76	0.890	0.960	18	Low	0.869	0.869	0.878	0.861	0.964	0.964
SqueezeNet	CNN	0.74	1.24	8.90	1.64	0.880	0.950	15	Moderate	0.874	0.870	0.856	0.857	0.936	0.936
ResNet101V2	CNN	1.92	10.12	124.3	1.90	0.910	0.970	25	Moderate	0.892	0.863	0.836	0.837	0.962	0.962
DenseNet201	CNN	2.46	12.95	157.7	2.15	0.885	0.955	27	Low	0.872	0.872	0.799	0.794	0.935	0.935
VGG19	CNN	4.12	23.51	282.6	3.05	0.880	0.930	30	High	0.889	0.894	0.856	0.853	0.892	0.887
ConvNeXtTiny	CNN	8.59	51.18	493.9	8.07	0.850	0.920	60	Low	0.776	0.877	0.811	0.798	0.900	0.900
DeepViT	Transformer	2.67	54.62	643.1	2.59	0.860	0.930	24	High	0.856	0.852	0.779	0.777	0.854	0.854
LeViT	Transformer	0.37	8.46	103.2	4.89	0.880	0.950	20	Moderate	0.865	0.863	0.833	0.834	0.966	0.966
SwinTransformer	Transformer	8.51	48.84	588.0	6.81	0.800	0.890	45	High	0.832	0.836	0.732	0.733	0.730	0.709
ViTbase	Transformer	3.42	68.54	822.7	3.33	0.785	0.860	35	Low	0.829	0.826	0.757	0.751	0.568	0.550
MaxViTsmall	Transformer	10.43	64.79	770.4	10.0	0.920	0.970	70	High	0.891	0.903	0.872	0.879	0.901	0.907
Proposed model	Hybrid	0.50	66.24	813.2	3.09	0.970	0.990	12	High	0.953	0.971	0.976	0.978	0.979	0.971

Quick links

Search