现代电子技术2024，Vol.47Issue(15)：60-64,5.DOI:10.16652/j.issn.1004-373x.2024.15.010
基于轻量级卷积神经网络的多视觉特征图像分割研究
Research on multi-visual feature image segmentation based on lightweight convolutional neural networks
摘要
Abstract
A multi-visual feature image segmentation method based on lightweight convolutional neural networks(CNNs)is studied to adapt to resource-constrained environments and meet the requirement of real-time performance.A lightweight multi-visual feature image segmentation model is designed based on the Linknet.The original multi-visual feature images are taken as encoder inputs.After preliminary feature extraction,the convolution kernels with different scales are used to learn the color,texture and other features in the multi-scale feature extraction module.In the channel attention module,squeeze-and-excitation block(SE block)is used to redirect the features with different scales.The feature extraction module A,which introduces depthwise separable convolution,is used to learn more abstract feature representations.The decoder uses feature extraction module B,deconvolution layer and standard convolution layer to transform the feature representations extracted by the encoder to generate feature maps containing semantic information.The bidirectional feature pyramid network is used to fuse the encoder and decoder to output features.The Sigmoid function is used to obtain multi-visual feature image segmentation results.The experimental results show that the training loss of the method studied is only 0.08,the method can achieve accurate segmentation of multi-visual feature images with MIoU(mean intersection over union)and F1-score of 0.912 8 and 0.906 8,respectively,and the parameter quantity,computational complexity and storage space of the segmentation model are 6.14 MB,1.52 GMac and 0.146 GB,respectively,so the method meets the lightweight requirements.关键词
轻量级/多视觉特征/图像分割/通道注意力/反卷积/双向特征金字塔Key words
lightweight/multi-visual feature/image segmentation/channel attention/deconvolution/bidirectional feature pyramid分类
电子信息工程引用本文复制引用
陈攀,王绍东..基于轻量级卷积神经网络的多视觉特征图像分割研究[J].现代电子技术,2024,47(15):60-64,5.基金项目
内蒙古自治区社会科学基金项目(2024DY05) （2024DY05）
内蒙古师范大学基本科研业务费专项资金资助(2023JBQN034) （2023JBQN034）
