HRRSI Building Extraction Method Based on Improved Unet and ConvCRF
Aiming at the problem of poor building edge segmentation caused by complex scenes in high-resolution remote sensing images,an improved building extraction method combining Unet and ConvCRF is proposed.The ordi-nary convolution operation in unet is replaced by the residual-block residual convolution structure,and Convolution Block Attention Module(CBAM)convolutional attention module is introduced in the low receptive field encoding and decoding stage of Unet to improve the model's ability to deal with building edges.extraction accuracy.At the same time,the ConvCRF module is connected to train the separation model to reduce the generation of jagged edges in the segmentation results,eliminate noise,and fit the real outline of the building.The experimental results show that the improved Unet neural network is superior to the classical semantic segmentation algorithm in segmentation effect and accuracy;the ConvCRF separation model can effectively eliminate noise and Reduces edge jaggedness.
BuildingsRemote sensing imagesDeep learningConditional random fields
万方数据
维普
