基于改进Faster-RCNN的目标检测算法研究

目标检测是图像处理领域一个重要的研究方向,深度学习方法需要大量数据进行训练,训练的繁杂和复杂的网络结构限制了目标检测的速度。本文基于Faster RCNN的网络架构,创新性提出了light tail Faster RCNN网络架构。light tail Faster RCNN算法在保证精度的情况下,大大提升了处理速度。在本文的设计中,通过将网络结构中的全连接层改为1*1的卷积层,来达到速度的提升。本文实验在PASCAL VOC数据集上进行,较经典网络模型,在识别率略低的情况下,速率提升了一倍多。在总体性能上显著优于经典目标检测算法,通过对比实验的方法比较验证了本文提出方法的有效性。

Target detection is an important research direction in the field of image processing.Deep learning methods require a large amount of data for training,and the complex and complex network structure of training limits the speed of target detection.Based on the network architecture of Faster RCNN,this paper innovatively proposes the light tail Faster RCNN network architecture.The Light tail Faster RCNN algorithm greatly improves the processing speed while ensuring accuracy.In the design of this article,the speed is improved by changing the fully connected layer in the network structure to a 1*1 convolutional layer.The experiment in this article is carried out on the PASCAL VOC data set.Compared with the classic network model,the speed is more than doubled when the recognition rate is slightly lower.The overall performance is significantly better than the classic target detection algorithm.The method comparison of the comparative experiment verifies the effectiveness of the method proposed in this paper.