| Autor: |
Weiye Sun, Yong Lin, Yezhi Wu, Lei Wang |
| Rok vydání: |
2021 |
| Předmět: |
|
| Zdroj: |
2021 IEEE 4th Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC). |
| DOI: |
10.1109/imcec51613.2021.9482098 |
| Popis: |
In order to salve the problem that it is difficult to detect the face objects wearing masks in the natural environments, a detection method based on improved Faster-RCNN was proposed. Based on the original Faster-RCNN as the overall framework, the ResNet-50 combined with the feature pyramid network (FPN) was used to replace the VGG-16 for feature extraction. K-Means algorithm and Soft-NMS algorithm were used to optimize the anchors to locate the targets efficiently. At the same time, imbalance problem between positive and negative samples was solved by introducing the Focal Loss function. Self-made face wearing diverse masks dataset were made to train the network and evaluate the network. The results of the experiment show that, comparing with the original networks, the mean average precision (mAP) of the optimized network improves by 6.4%. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
