TABLE 3.

Mean average precision (mAP), frames per second (FPS), root mean square error (RMSE), and root mean square percentage error (RMSPE) of SSD, Yolov3, Yolov4, Yolov5s, Yolov5m, Faster R-CNN, and our method in detecting wheat spikes.

Method	Backbone	Datasets	mAP50	mAP75	FPS	RMSE	RMSPE	Counting Acc
SSD	VGG	Mixed	0.4356	0.1652	60	10.30	0.26	0.4841
Yolov3	DarkNet53 + FPN	Mixed	0.8983	0.4832	50	2.56	0.08	0.8991
Yolov4	CSPDarkNet53 + PANet	Mixed	0.9127	0.4902	52	2.13	0.14	0.9095
Yolov5s	CSPDarkNet53 + PANet	Mixed	0.9272	0.5128	60	1.71	0.12	0.9302
Yolov5m	CSPDarkNet53 + PANet	Mixed	0.9312	0.5217	50	1.53	0.06	0.9330
Faster-RCNN	ResNet101 + FPN	Mixed	0.8536	0.4956	10	3.14	0.07	0.8805
Our method	ResNet101 + BiFPN	Mixed	0.9262	0.5023	22	1.96	0.06	0.9288

The results of our method are highlighted in bold.