Road Extraction Method Oriented by Multi-Scale and Strip Features
In the task of extracting roads from remote sensing images,road information is often affected by environmental factors such as light-ing,shadows,and occlusion,and roads usually appear as slender strips,making it difficult to accurately detect.To this end,an improved LinkNet model(MSS LinkNet)for multi-scale and strip features is proposed to capture contextual information at different scales,which is highly compatible with the slender characteristics of roads.Firstly,the multi-scale convolutional attention module is used as the basic compo-nent unit of the encoder to ensure the model's ability to extract multi-scale and stripe features.Secondly,an improved hollow space pyramid pooling module is added to the central area of the network to enhance the model's ability to parse multi-scale information.Finally,a bar pool-ing module is added to the decoder section to enhance the model's ability to parse bar information.The experiment shows that compared to D-LinkNet,the proposed model has improved IOU by 2.53%and 0.71%on the DeepGlobe and Massachusetts datasets,respectively,while only accounting for 54.15%and 79.63%of D-LinkNet in terms of parameter and computational complexity.
国家科技期刊平台
NSTL
万方数据
