基于深度学习多分类模型的模糊图像目标形态提取方法
An Extraction Method of Fuzzy Image Target Morphology Based on Deep Learning Multi Classification Model
荣蓉1
作者信息
- 1. 山东工艺美术学院公共课教学部,济南 250300
- 折叠
摘要
目前常规的模糊图像目标形态提取方法主要通过结合自编码器对图像轮廓边界进行识别,由于在对图像进行滤波的过程中丢失了较多的边缘特征,导致目标形态提取效果不佳.对此,提出基于深度学习多分类模型的模糊图像目标形态提取方法.首先,对模糊图像进行各向异性扩散滤波处理,调整扩散系数,从而保证模糊图像在去除斑点噪声的同时能够保留更多的边缘特征信息.其次结合能量函数分类像素归类的惩罚概率代价,实现图像分割处理.最后采用卷积层对输入的样本特征数据进行融合处理,从而实现目标形态特征提取.实验结果表明,采用提出的方法对模糊图像进行目标形态提取时,图像的惯性矩较高,平均为0.624,具有较为理想的形态提取效果.
Abstract
At present,the conventional methods for extracting target morphology from fuzzy images mainly recognize image contour boundaries by combining autoencoders.However,due to the loss of many edge features during the filtering process of the image,the effect of target morphology extraction is not satisfactory.To address this issue,a fuzzy image target morphology extraction method based on deep learning multi classification model is proposed.Firstly,anisotropic diffusion filtering is applied to the fuzzy image to adjust the diffusion coefficient,ensuring that the fuzzy image retains more edge feature information while removing speckle noise.Secondly,combined with the energy function,the punishment probability cost of pixel classification is classified to achieve image segmentation processing.Finally,convolutional layers are used to fuse the input sample feature data to achieve the extraction of target morphological features.The experimental results show that when using the proposed method to extract target morphology from fuzzy images,the inertia moment of the image is relatively high,with an average of 0.624,which has an ideal morphology extraction effect.
关键词
深度学习/模糊图像/形态提取/滤波降噪Key words
deep learning/fuzzy image/morphology extraction/filtering and noise reduction引用本文复制引用
出版年
2024
NETL
NSTL
万方数据