基于全卷积神经网络复杂场景的车辆分割研究

Research on Vehicle Segmentation Based on Complex Scenes of Full Convolutional Neural Network

针对目前存在的复杂交通场景中车辆分割精度不足的问题,本文提出了一种基于全卷积神经网络对图像中车辆进行分割的方法。在VGG16Net基础上,将全连接层改为卷积层,为获得更精细的边缘分类结果,减少了部分卷积层,并融合浅层和深层特征,同时,为提高交通环境下车辆的分割精度,减少其他类别目标的干扰,将对车辆目标的分割问题改为基于像素的二分类问题,为提高网络的训练速度,采用Adam优化算法对网络进行训练。实验结果表明,与现有的全卷积神经网络分割效果相比,该网络对复杂交通场景下的车辆分割精度明显提高。该研究在智能交通方面具有较好的应用前景。

Aiming at the problem of insufficient vehicle segmentation accuracy in the existing complex traffic scenes, this paper proposes a full convolutional neural network to realize the segmentation of vehicles in the image. On the basis of VGG16Net, the full connection layer is changed to the convolution layer, and in order to obtain finer edge classification results, the partial convolution layer is reduced and the shallow deep features are merged;In order to improve the segmentation accuracy of vehicles in the traffic environment and reduce the interference of other categories of targets, the problem of segmentation of vehicle targets is changed to the pixel-based classification problem;In order to improve the training speed of the network, the Adam optimization algorithm is used to train the network. Finally, the experimental results show that compared with the existing full convolutional neural network segmentation effect, the network has significantly improved vehicle segmentation accuracy under complex traffic scenarios. This research has a good application prospect in intelligent transportation.