| 摘要: |
| 针对SSD(Single Shot MultiBox Detector)目标检测算法对小目标检测能力不足的问题，提出一种引入视觉机制和多尺度语义信息融合的VFF-SSD(Vision Feature Fusion SSD)改进算法。为了增大浅层网络的感受野提高特征提取能力，首先在SSD浅层特征层中加入视觉机制，然后利用改进PANet(Path Aggregation Network)多尺度特征融合网络与深层特征增强网络得到新的特征层，旨在增强浅层网络的语义信息并加强深层特征的特征表达能力，最后应用注意力机制模块提高对重要信息的学习能力。实验结果表明，在PASCAL VOC2007测试集检测的mAP(Mean Average Precision)值达到81.1%，对数据集中小目标的mAP值较原SSD提高了6.6%。 |
| 关键词: 小目标检测 深度学习 视觉机制 多尺度语义信息 注意力机制 |
| DOI：10.20079/j.issn.1001-893x.220901003 |
|
| 基金项目:辽宁省自然科学基金（20180520022） |
|
| A Small Object Detection Algorithm Fusing Vision Mechanism and Multi-scale Features |
| WU Debin,LIU Xiaonan,LIU Zhenyu,YANG Na |
| (School of Information Science and Engineering，Shenyang University of Technology，Shenyang 110870，China) |
| Abstract: |
| For the problem that the Single Shot Multibox Detector(SSD) object detection algorithm has insufficient ability to detect small targets,an improved Vision Feature Fusion SSD(VFF-SSD) algorithm is proposed in which visual mechanism and multi-scale semantic information fusion are adopted.In order to increase the receptive field of the shallow network and improve the feature extraction ability,a visual mechanism is added to the SSD shallow feature layer.Then,new feature layers is obtained by using the improved path aggregation network(PANet) multi-scale feature fusion network and the deep feature enhancement network,to enhance the semantic information of the shallow network and enhance the feature expression ability of the deep features.Finally,the attention mechanism module is applied to improve the learning ability of important information.The experimental results show that the mean average precision(mAP) value detected in the PASCAL VOC2007 test set reaches 81.1%,and the mAP value for small targets in the data set is 6.6% higher than that of the original SSD. |
| Key words: small object detection deep learning vision mechanism multi-scale semantic information attention mechanism |
