Deep learning-driven image segmentation:a comparative study of U-Net,SegNet and DenseASPP
The study aimed to thoroughly compare deep learning-driven image segmentation methods,particularly focusing on the performance differences among three classic architectures:U-Net,SegNet,and DenseASPP.The authors conducted extensive experiments on the crack dataset,fire dataset,and Cityscapes dataset,comparing the models in terms of accuracy,efficiency,and robustness,providing a comprehensive assessment of their performance in image segmentation tasks.The results indicated that Seg-Net slightly lagged behind U-Net on the crack dataset,but DenseASPP consistently outperformed on all datasets,demonstrating higher accuracy and robustness.This study highlights the crucial role of model architecture in image segmentation performance,making DenseASPP a promising choice for tasks requiring high precision and robustness.The research offers valuable guidance for deep learning image segmentation tasks,helping researchers make informed decisions regarding the selection of model architec-tures tailored to their specific tasks and providing valuable insights for the future of the field of image segmentation.
万方数据
维普
