基于改进HED的多尺度边缘检测方法

Improved multi-scale edge detection method based on HED

针对基于深度学习的多尺度边缘检测不可避免出现自适应性低,参数增加,计算量大,检测边缘不连续等问题,本文提出一种基于改进整体嵌套的多尺度边缘检测方法.该方法将多尺度检测与弱监督模型相结合,解决参数多计算量大的问题.为了充分利用卷积强大的特征表达能力,在整体嵌套边缘检测的基础上,提出了一种多尺度下深度学习结构,一个相互独立的多网络多尺度结构,由不同深度和输出的多个网络组合.同时引用整体嵌套的权重混合层,权重混合层将所有的弱监督预测结果连接到一起,并在训练的过程中学习混合权重。将所提方法在数据集BSDS500上通过评价指标进行性能评价.实验结果表明,所提方法能够在数据集BSDS500上达到很好的性能.

Aiming at the problems of deep learning-based multi-scale edge detection inevitably low adaptability,increased parameters,large calculations,and discontinuous detection edges,this paper proposes a multi-scale edge detection method based on improved overall nesting.The method combines multi-scale detection with weak supervision model to solve the problem of large amount of parameter calculation.In order to make full use of the powerful feature expression ability of convolution,based on the whole nested edge detection,a multi-scale deep learning structure is proposed,which is an independent multi-network multi-scale structure,which is composed of multiple networks with different depths and outputs.At the same time,the whole nested weight mixing layer is referenced.The weight mixing layer connects all the weak supervised prediction results together and learns the mixed weight in the training process.The performance of the proposed method is evaluated by the evaluation index on the data set BSDS500.Experimental results show that the proposed method can achieve good performance on BSDS500.