摘要
针对基于卷积神经网络的图像边缘检测算法忽略中间层而丢失图像细节信息的问题,提出一种基于充分卷积的边缘检测算法。在视觉几何组16骨干网络上剪切掉所有的全连接层和池化层以构建全卷积网络;在全卷积网络每个阶段的1×1×21卷积层后边连接累加层获取每个阶段中的特征信息;通过融合层替换原来位置剪切掉的全连接层,在融合层的后边连接卷积层和损失函数层,并重新计算损失函数训练网络参数。实验结果表明,该算法的边缘检测性能优于人眼的平均性能,也比其他边缘检测算法性能更优。
To deal with the problem of ignoring the middle layer and losing the details of the image when detecting the edge of the image with the convolutional neural network method,an edge detection algorithm based on full convolution is proposed.The all fully connected layers and pooling layers are cut on the 16-backbone network of the visual geometry group to build a fully convolutional network;an accumulation layer behind the convolutional layer at each stage of the fully convolutional network is connected to obtain the feature information;the fully connected layer that was cut out at the original position is replaced with the fusion layer,the convolution layer and the loss function layer are connected behind the fusion layer,and the loss function training network parameters are recalculated.Experimental results show that the edge detection performance of this algorithm is better than that of human visual perception and is also better than other edge detection algorithms.
作者
谢晓飞
来毅
刘颖
XIE Xiaofei;LAI Yi;LIU Ying(Center for Image and Information Processing,Xi'an University of Posts and Telecommunications,Xi'an 710121,China;Key Laboratory of Electronic Information Application Technology for Scene Investigation,Ministry of Public Security,Xi'an University of Posts and Telecommunications,Xi'an 710121,China)
出处
《西安邮电大学学报》
2020年第5期50-54,共5页
Journal of Xi’an University of Posts and Telecommunications
关键词
深度学习
卷积神经网络
边缘检测
特征提取
deep learning
convolutional neural network
edge detection
feature extraction