Research on Image Segmentation Methods Based on Convolutional Neural Networks
Aiming at the problems of multiple parameters,overfitting leading to low image segmentation ac-curacy and low algorithm efficiency in traditional convolutional neural networks,maximum pooling pro-cessing is adopted to replace the downsampling layer,and an improved CNN structure is constructed to ob-tain the U-Net convolutional neural network,which is further improved.The improved U-Net convolution-al neural network is applied to high-resolution remote sensing images,and the results show that it can per-form fine and complete segmentation of small buildings in remote sensing images.In addition,by compa-ring with FCN32s,SegNet,and FCN8s,it is pointed out that the improved U-Net convolutional neural net-work has better performance in remote sensing image segmentation.
国家科技期刊平台
NSTL
万方数据
