5 Experimental results and analysis
⑴ mAP value comparison
In this paper, the performance of Faster RCNN, RetinaNet and improved
RetinaNet are compared. The model comparison results are shown in Table
3. The mAP values for backbone of Faster RCNN using vgg, resnet and
adding CBAM are 78.39%, 77.71% and 78.41%. The mAP values of
RetinaNet, RetinaNet with SE, CBAM and improved RetinaNet are 89.74%,
89.69%, 90.80% and 91.84%, respectively. The results show that the
mAP values of the improved RetinaNet are improved by 13.45% and 2.1%,
respectively, compared to the Faster RCNN and RetinaNet.
Tab 3 mAP value comparison of models