基于深度卷积神经网络的道路场景理解

发布时间：2018-07-05 03:35

本文选题：深度学习 + 卷积神经网络　；参考：《计算机工程与应用》2017年22期

【摘要】：在无人驾驶技术中,道路场景的理解是一个非常重要的环境感知任务,也是一个很具有挑战性的课题。提出了一个深层的道路场景分割网络(Road Scene Segmentation Network,RSSNet),该网络为32层的全卷积神经网络,由卷积编码网络和反卷积解码网络组成。网络中采用批正则化层防止了深度网络在训练中容易出现的"梯度消失"问题;在激活层中采用了Maxout激活函数,进一步缓解了梯度消失,避免网络陷入饱和模式以及出现神经元死亡现象;同时在网络中适当使用Dropout操作,防止了模型出现过拟合现象;编码网络存储了特征图的最大池化索引并在解码网络中使用它们,保留了重要的边缘信息。实验证明,该网络能够大大提高训练效率和分割精度,有效识别道路场景图像中各像素的类别并对目标进行平滑分割,为无人驾驶汽车提供有价值的道路环境信息。
[Abstract]:In driverless technology, the understanding of road scene is a very important task of environmental awareness, and also a very challenging task. A deep road scene segmentation network (RSSNet) is proposed. The network is a 32-layer full-convolution neural network composed of convolutional coding network and deconvolution decoding network. In the network, batch regularization layer is used to prevent the problem of "gradient disappearing", and Maxout activation function is used in the activation layer, which further alleviates the gradient disappearance. At the same time, the Dropout operation is used properly in the network to prevent the over-fitting of the model. The coding network stores the largest pooled index of the feature graph and uses them in the decoding network, thus preserving important edge information. Experiments show that the network can greatly improve the training efficiency and segmentation accuracy, effectively identify the categories of each pixel in the road scene image and smooth the target segmentation, and provide valuable road environment information for driverless vehicles.
【作者单位】：咸阳师范学院计算机学院;西安理工大学机械与精密仪器工程学院;
【基金】：国家自然科学基金(No.10872160)
【分类号】：TP183;TP391.41

【相似文献】