TABLE 4.

Detection results of various CNN models.

Method	SSD	R-FCN	Faster R-CNN						Faster DR-IACNN
Feature extractor	VGG 16	ResNet 50	ZF	VGG 16	ResNet 50	ResNet 34	ResNet 18	INSE-ResNet (our work)	INSE-ResNet (our work)
Classes
Input	512	500	500	500	500	500	500	500	500
Iterations	120 k	200 k	200 k	200 k	200 k	200 k	200 k	280 k	280 k
Black rot	74.7	79.0	63.5	64.5	64.4	69.3	65.8	73.7	76.7
Black measles	81.6	82.5	75.4	79.9	79.0	81.4	75.0	85.3	88.0
Leaf blight	72.0	68.2	59.2	60.0	60.8	64.4	64.4	71.1	73.7
Mites of grape	77.9	69.4	69.3	70.4	70.2	70.8	73.6	84.0	86.2
mAP (%)	76.6	74.8	66.9	68.7	68.6	71.5	69.7	78.5	81.1

The meaning of the “bold” is the best experimental results.