• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>IEEE Joint International Information Technology and Artificial Intelligence Conference >General Target Detection Method Based on Improved SSD
【24h】

General Target Detection Method Based on Improved SSD

机译：基于改进SSD的通用目标检测方法

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In order to improve the accuracy of general target detection by the surveillance camera in the monitoring area, an improved SSD model named DC-SSD that integrates the deformable convolution network was proposed. This model is based on the SSD convolutional neural network model, and the idea of deformable convolution is used for reference. The deformable convolution module is added into the three high level convolution modules of SSD to make the sampling point position of the high level convolution kernel adaptively change according to the image content, so as to adapt to the geometric deformation of different objects such as shape and size. In the open source neural network framework Caffe, DC-SSD was tested with PASCAL VOC 2007 (train+val) as the training set and PASCAL VOC 2007 (test) as the test set. Under the same training and test conditions, the mAP of DC-SSD reached 72.0%, 4.0% higher than the original SSD model, and the detection frame rate reached 39fps. Experimental results show that DC-SSD model can effectively improve the accuracy of universal target detection and meet the real-time requirements.
机译：为了提高监视摄像机在监视区域内对普通目标的探测精度，提出了一种改进的DC-SSD固态硬盘模型，该模型集成了可变形卷积网络。该模型基于SSD卷积神经网络模型，并以可变形卷积的思想为参考。将可变形卷积模块添加到SSD的三个高级卷积模块中，使高级卷积核的采样点位置根据图像内容自适应地变化，以适应形状和形状等不同对象的几何变形。尺寸。在开源神经网络框架Caffe中，以PASCAL VOC 2007（train + val）作为训练集，以PASCAL VOC 2007（test）作为测试集对DC-SSD进行了测试。在相同的训练和测试条件下，DC-SSD的mAP达到72.0％，比原始SSD模型高4.0％，检测帧率达到39fps。实验结果表明，DC-SSD模型可以有效提高通用目标检测的准确性，并满足实时性要求。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号