Building extraction from remote sensing images based on lightweight network
Existing building extraction algorithms focus on the improvement of accuracy while ignoring the increase in model computation amount and parameters.To address this issue,a lightweight network model for building extraction from high-resolution remote sensing images was designed.The model was based on the U-Net structure,and a hybrid convolutional unit consisting of depthwise separable convolution and ordinary convolution was used to build the model,so as to reduce the computation amount and parameters of the model.At the same time,a lightweight dual attention module was added behind each unit of the model to enhance the feature extraction capability of the model and improve the building extraction accuracy,realizing a balance between performance and spatiotemporal complexity.The experimental results on Satellite dataset Ⅱ datasets show that the intersection over union(IoU)and F1 score of the lightweight network model reach 0.696 4 and 0.821 1,which are 4.45%and 3.18%higher than those of the U-Net model,respectively.The amount of computation and parameters are reduced by 34.56%and 44.79%compared with those of the U-Net model,which results in a significant improvement in overall performance.In terms of extraction effect,the model has better extraction results than other neural network models when facing the interference of complex backgrounds,small buildings,and surrounding features.
国家科技期刊平台
NSTL
万方数据
