改进U-Net网络的接地网图像超像素分割北大核心CSCD

研究基于改进U-Net网络的接地网图像超像素分割方法,提升红外图像超像素分割效果。通过主成分分析法降维处理接地网腐蚀红外图像;利用Turbopixel超像素分割法分割降维后的红外图像,获取数个超像素区域;在全卷积U-Net网络内添加可变形卷积与重构上采样卷积,并利用反向传播算法,优化网络参数,建立改进的全卷积U-Net网络结构;在改进的全卷积U-Net网络内分割获取的数个超像素区域,输出红外图像超像素自动分割结果。实验证明:该方法可有效降维处理接地网腐蚀红外图像,实现红外图像超像素分割,分割后的红外图像边界清晰;在不同分辨率时,该方法的Dice相似性系数较高、Hausdorff距离较低,具备较高的红外图像超像素分割精度。

Improved ground network image superpixelsegmentation for U Net network

The super pixel segmentation method of grounding grid image based on improved U Net network is studied to improve the super pixel segmentation effect of infrared image.Principal component analysis is used to reduce the dimension and process the infrared image of grounding grid corrosion.The infrared image after dimension reduction is segmented by Turbopixel superpixel segmentation method,and several superpixel regions are obtained.The deformable convolution and reconstructed upsampling convolution are added to the full convolution U NET network,and the network parameters are optimized by using the back propagation algorithm to establish an improved full convolution U NET network structure.Several superpixel regions obtained by segmentation are input in the improved fully convolution U NET network,and the automatic segmentation results of infrared image superpixel are output.The experimental results show that the proposed method can effectively reduce the dimension of the ground grid corrosion infrared image and realize the superpixel segmentation of the infrared image,and the boundary of the segmented infrared image is clear.At different resolutions,the proposed method has higher Dice similarity coefficient and lower Hausdorff distance,and has higher segmentation accuracy of infrared image superpixels.