Table 2 An ablation study evaluating the proposed model using standalone CNN backbones and their combinations with CA mechanisms for pest and disease recognition.

Method	Parameters (million)	Model Size (MB)	Precision	Recall	F1-score	Accuracy	Matthews correlation coefficient
VGG16⁵⁹	138.4	528	0.800	0.79	0.794	74.62	0.593
ResNet-50⁵⁵	25.6	98	0.806	0.812	0.808	76.91	0.617
DenseNet-121⁵⁸	8.1	33	0.818	0.800	0.808	76.84	0.622
Xception	22.9	88	0.762	0.730	0.745	73.50	0.502
InceptionV3	23.9	92	0.860	0.815	0.83.6	81.20	0.683
VGG16 + CA	140	540	0.820	0.805	0.812	77.80	0.628
ResNet-50 + CA	26	100	0.835	0.811	0.822	80.90	0.651
DenseNet-121 + CA	8.3	35	0.830	0.810	0.821	80.55	0644
Xception + CA	24.1	90	0.800	0.810	0.804	76.50	0.608
The proposed model	*7.9*	32	*0.932*	*0.891*	*0.911*	*88.50*	*0.816*

Quick links

Search