基于RGB-D图像和深度学习的场景语义分割网络

近年来，深度卷积神经网络应用于图像语义分割领域并取得了巨大成功。本文提出了一个基于RGB-D(彩色-深度) 图像的场景语义分割网络。该网络通过融合多级RGB网络特征图和深度图网络特征图，有效提高了卷积神经网络语义分割的准确率。同时，本文利用带孔的卷积核设计了具有捷径恒等连接的空间金字塔结构来提取高层次特征的多尺度信息。在SUN RGB-D数据集上的测试结果显示，与其它state-of-the-art的语义分割网络结构相比，本文所提出的场景语义分割网络性能突出。

Scene Parse Network based on RGB-D Image and Deep Learning

In recent years, deep convolutional neural networks have been applied to the field of image semantic segmentation and achieved great success. This paper presents a scene semantic segmentation network based on RGB-D(RedGreenBlue-Depth) images. The network effectively improves the accuracy of semantic segmentation of convolutional neural networks by merging multi-level RGB network features and depth network features. At the same time, this paper using convolution kernels with holes designs a spatial pyramid structure with shortcut to extract high-level features of multi-scale information. The test results on the SUN RGB-D dataset show that, compared with other state-of-the-art semantic segmentation networks, the performance of the semantic segmentation network we propose in this paper is outstanding.