DA-ActNN-YOLOV5: Hybrid YOLO v5 Model with Data Augmentation and Activation of Compression Mechanism for Potato Disease Identification

Table 6

mAP and training elapsed time of DA-ActNN-YOLOV5 on the PBD-IM dataset.

Model	Batch size	Image size	Level	mAP (%)	Time (h)
DA-ActNN-YOLOV5m	32	384	—	98.76	1.852
	64	384	—	99.86	1.579
	128	384	—	98.99	1.286
	64	512	—	99.68	1.685
	128	512	L1	OOM	—
	128	512	L3	98.66	1.663
	128	512	L5	99.12	1.681
DA-ActNN-YOLOV5x	16	384	—	97.23	2.869
	32	384	—	98.68	2.521
	64	384	—	99.78	2.219
	64	512	L1	OOM	—
	64	512	L3	97.99	2.674
	64	512	L5	98.39	2.683
DA-ActNN-YOLOV5l	8	384	—	96.89	3.853
	16	384	—	97.38	3.457
	32	384	—	98.97	3.289
	32	512	—	OOM	—
	32	512	L1	OOM	—
	32	512	L3	98.57	3.565
	32	512	L5	98.19	3.723