基于卷积神经网络的语义分割算法研究

针对语义分割中残差网络并不能完好地提取图像信息和分割效果差的问题,提出一种联合特征金字塔模型(JFP)用来融合残差网络的输出特征,并结合暗黑空间金字塔池化模型(ASPP)进一步提取特征。在解码部分应用简单的解码结构,恢复图像尺寸完成语义分割;同时引入注意力模型作为辅助语义分割网络,辅助神经网络进行训练。该方法分别在Pascal VOC 2012数据集和增强的Pascal VOC 2012数据集上对网络进行训练,并在Pascal VOC 2012的验证集上进行测试,其平均交并集之比(mIoU)分别达到了78.55%和80.14%,表明该方法具有良好的语义分割性能。

Research on semantic segmentation algorithm based on convolutional neural network

In order to solve the problem that the residual network cannot extract image information well and the segmentation effect is poor in semantic segmentation,this paper proposed a joint feature pyramid model to integrate the output features of the residual network,and then further extracted the features in combination with the atrous spatial pyramid pooling module.In the decoding part,this paper applied a simple decoding structure to recover the image size to complete the semantic segmentation.This paper also used attention module as the auxiliary semantic segmentation network to assist the training of the neural network.This method trained the network in the Pascal VOC 2012 data set and the enhanced Pascal VOC 2012 data set respectively,and tested it on the verification set of Pascal VOC 2012.The average ratio of intersection and union(mIoU)is 78.55%and 80.14%respectively,which shows that proposed method has good semantic segmentation performance.