Research on Small Target Detection Technology Based on the MPH-SSD Algorithm

Table 4

Performance comparison of the MPH-SSD algorithm and other algorithms on the MS COCO dataset.

Method	Backbone network	Avg. precision, IoU			Avg. precision, area			Avg. recall, area
		IoU = 0.5 : 0.95	IoU = 0.5	IoU = 0.75	Area : S	Area : M	Area : L	Area : S	Area : M	Area : L
Faster R-CNN [37]	VGG16	24.2	45.3	23.5	7.7	26.4	37.1	—	—	—
Mask R-CNN [29]	ResNeXt-101-FPN	37.1	60.0	39.4	16.9	39.9	53.5	—	—	—
YOLOv2 [40]	Darknet19	21.6	44.0	19.2	5.0	22.4	35.5	9.8	36.5	54.4
SSD512 [40]	VGG16	27.7	46.4	26.7	10.9	31.8	43.5	16.5	46.6	60.8
DSSD513 [41]	ResNet-101	33.2	53.3	35.2	13.0	35.4	51.1	28.9	43.5	46.2
DF-SSD [43]	DenseNet-S-32-1	29.5	50.7	31.3	9.8	31.1	46.5	17.3	46.8	64.4
SEFN512 [44]	VGG16	33.7	54.7	35.6	19.2	38.0	47.3	29.1	52.5	63.2
FSSD512 [43]	VGG16	31.8	52.8	33.5	14.2	35.1	45.0	22.3	49.9	62.0
RFB512 [44]	VGG16	34.4	55.7	36.4	17.6	37.0	47.6	27.3	52.3	65.4
MPH-SSD	VGG16	51.1	79.8	55.6	21.1	36.3	59.4	28.4	46.8	67.9