Table 3.

Comparison summary of pre-trained models vs. proposed architecture on banana leaf disease dataset (Accuracy: Acc, Precision: P, Recall: R, Average time per epoch in seconds: ATPES, Average GPU Inference Time in Seconds: AGITS).

Model	Acc	P	R	AUC	ATPES	AGITS
MobileNetV2	0.7531	0.7660	0.7407	0.9104	10	0.088
DenseNet121	0.8148	0.8174	0.8107	0.9251	15.1	0.096
ResNet152V2	0.7942	0.8042	0.7942	0.9180	19.3	0.129
DenseNet169	0.8107	0.8133	0.8066	0.9469	15.7	0.097
DenseNet201	0.8230	0.8257	0.8189	0.9429	17.3	0.098
InceptionV3	0.7407	0.7586	0.7243	0.8983	20.7	0.142
NASNetLarge	0.7860	0.7925	0.7860	0.9184	27.9	0.163
InceptionResNetV2	0.6955	0.7294	0.6543	0.8275	22.8	0.154
EfficientNetV2S	0.7819	0.7908	0.7778	0.9313	16.02	0.107
EfficientNetV2L	0.7407	0.7469	0.7407	0.9059	32.2	0.198
Modified DenseNet201_PReLU	0.8971	0.9004	0.8930	0.9716	17.37	0.098
Modified DenseNet201_Relu	0.8971	0.8996	0.8848	0.9670	17.37	0.098
DenseNet201Plus with only attention mechanism	0.8712	0.8501	0.8501	0.9529	17.37	0.098
DenseNet201Plus with only attentive transition	0.8788	0.8694	0.8714	0.9602	17.37	0.098
Proposed DenseNet201Plus	0.9012	0.9012	0.9012	0.9716	17.37	0.098