Detecting Citrus in Orchard Environment by Using Improved YOLOv4

Table 5

Identification parameters of four methods for different occlusion degrees.

Occlusion condition	Model	Citrus count	Correctly identified		Falsely identified		Missed
			Amount	Rate (%)	Amount	Rate (%)	Amount	Rate (%)
Less than 50%	Faster-RCNN	200	162	81.24	19	9.71	31	15.44
	YOLOv3	200	156	78.22	23	11.52	38	18.96
	YOLOv4	200	173	86.38	14	7.10	22	11.21
	Improved YOLOv4	200	187	93.58	12	5.98	16	8.15
More than 50%	Faster-RCNN	200	156	78.24	26	12.81	39	19.34
	YOLOv3	200	148	74.22	35	17.52	46	22.96
	YOLOv4	200	166	83.18	20	10.05	26	13.07
	Improved YOLOv4	200	182	90.82	14	7.18	21	10.36