融合FCN和ECA的FE-CCNN图像语义分割算法OA
FE-CCNN Image Semantic Segmentation Algorithm Fusing FCN and ECA
为了解决图像语义分割任务中分割不连续、误分割、物体边缘分割困难等问题,提出一种融合全卷积网络(fully convolutional network,FCN)和高效注意力机制(efficient channel attention,ECA)的 FE-CCNN 语义分割算法.首先,该算法采用 ResNet50 作为 FCN 模型的骨干网络,用于图像特征的提取.其次,使用连续耦合神经网络(con-tinuous-coupled neural network,CCNN)模型提升特征提取能力,再通过 ECA 注意力机制对特征通道进行加权处理.最后,将特征图上采样至原图尺寸并输出分割结果.所提算法在 PASCAL_VOC2012 数据集和 CamVid 数据集上进行了实验,其像素精度分别达到了 93.5%和 89.7%,平均 IoU 分别达到了 71.2%和 70.2%,均优于其他对比算法.此外,消融实验进一步验证了 FE-CCNN 算法在图像语义分割任务中的有效性.
To address the issues of discontinuous segmentation,incorrect segmentation,and the difficulty of segmenting object edges in image semantic segmentation tasks,a FE-CCNN algorithm for image seman-tic segmentation that integrated fully convolutional network(FCN)and efficient channel attention(ECA)was proposed.Firstly,ResNet-50 was adopted as the backbone network of the FCN model to extract im-age features.Secondly,the cascaded convolutional neural network(CCNN)was introduced to enhance the capability of image feature extraction,followed by the application of the ECA attention mechanism to weight the importance of feature channels.Finally,the the feature maps were upsampled to the original image size to output the segmentation results.The proposed algorithm was tested on the PASCAL_VOC 2012 dataset and the CamVid dataset,achieving pixel accuracy(PA)of 93.5%and 89.7%,and mean intersection over union(mIoU)scores of 71.2%and 70.2%,respectively,which were superior to those of other comparative algorithms.Moreover,via ablation studies,the effectiveness of the FE-CCNN algo-rithm in image semantic segmentation tasks was further verified.
刘欢;张莉;安新磊;常举越
兰州交通大学 数理学院 甘肃 兰州 730070兰州工业学院 基础学科部 甘肃 兰州 730050兰州交通大学 数理学院 甘肃 兰州 730070兰州交通大学 数理学院 甘肃 兰州 730070
信息技术与安全科学
图像语义分割全卷积网络连续耦合神经网络高效注意力机制
image semantic segmentationfully convolutional networkcontinuous-coupled neural net-workefficient channel attention
《郑州大学学报(理学版)》 2026 (5)
54-61,8
甘肃省自然科学基金项目(23JRRA861,24JRRA291)兰州工业学院青年科技创新项目(2021KJ-08)
评论