Complex scene segmentation of collapsed buildings algorithm based on improved DeepLabV3+
Existing segmentation methods for post-earthquake collapsed buildings suffer from low accuracy and real-time pro-cessing challenges due to factors like backlighting,thermal noise,and rain and fog interference.In response to these issues,a method is proposed for segmenting collapsed buildings in complex scenarios based on DeepLabV3+.Firstly,we select the Mobile-NetV2 backbone network as the feature extractor for DeepLabV3+to reduce network computational parameters and address the slow segmentation speed.Subsequently,an AS block is introduced in the feature fusion stage to selectively enhance feature channel weight coefficients related to collapsed buildings,thereby improving the model's segmentation accuracy.Additionally,within the backbone network,FPN is introduced to integrate multi-level features in the decoding stage,effectively capturing information from various collapsed building ranges.The experimental findings indicate that this network has,to a certain extent,effectively tackled the challenges related to segmentation efficiency and accuracy in intricate scenarios involving collapsed structures.
deep learningimage segmentationcollapsed buildingscomplex scenes
NETL
NSTL
万方数据
