基于区域生成网络结构的多层特征融合目标检测算法

Multi-feature Fusion Object Detection Algorithm Based on Region Proposal Network Structure

现有深度学习目标检测算法往往只利用了卷积神经网络(convolutional neural network,CNN)提取的深层特征进行判别,对浅层特征利用不足。为了利用浅层的细节信息来提高最终所提取的特征层信息的丰富性,提出了一种基于区域生成网络(region proposal network,RPN)结构的多层特征融合目标检测算法,该算法通过深度卷积网络获取不同层次的特征,并将浅层特征与深层次特征进行融合来获得更加丰富的提取特征,以提升检测模型的性能。以ImageNet上的公开数据voc2007为实验对象,以Faster RCNN为基础的检测框架进行改进,最终改进后的平均精度均值(mean average precision,mAP)相比于Faster RCNN有所提升,表明研究结果提升了目标检测模型的准确度。

The current objection detection algorithm usually only use the deep feature of the convolutional neural network(CNN)feature extraction network to detect the object,and it can t make good use of shallow feature.In order to utilize the shallow details to improve the richness of the final extracted feature layer information,a multi-feature fusion object detection algorithm was presented based on region proposal network(RPN)structure that get different layer feature by deep convolutional neural network and made these shallow feature merge with these deep feature to improve the richness of the final extracted feature layer information and to improve the performance of the detection model.the public data set voc2007 from the ImageNet were used as train and test subject,and the detection framework based on Faster RCNN was improved,and the final mean average precision(mAP)was improved compared to the mAP of Faster RCNN.The final result shows that this algorithm improves the accuracy of the object detection model.