基于CBAM和Unet的遥感影像水体识别

Water Body Recognition by Remote Sensing Images based on CBAM and Unet

使用Unet深度学习技术,引入注意力机制CBAM(Convolutional Block Attention Module)动态捕捉图像的关键特征信息,并根据每个通道的重要性自适应地调整注意力权重,增强水体识别模型的表达能力和性能。通过实验验证,相比Unet水体识别模型,CBAM+Unet水体识别模型识别的河流在宽度、走向、轮廓上更接近真实河流,而且对河流的边线识别也更加精细,该模型的准确率、精确率、召回率、F1值、Kappa系数各项指标分别达到98.24%、98.73%、99.32%、99.02%、89.77%,Kappa系数和Unet相比提高8.52%,说明CBAM+Unet水体识别模型具有更高的识别精度和水边线提取能力。

In order to solve these problems,this paper used Unet deep learning technology to introduce the attention mechanism CBAM(Convolutional Block Attention Module)to dynamically capture the key feature information of the image,and adaptively adjust the attention weight according to the importance of each channel to enhance the expressive ability and performance of the water body recognition model.Through experimental verification,compared with the Unet water body recognition model,the river recognized by the CBAM+Unet water body recognition model was closer to the real river in width and direction and contour.The river edge recognition was much better.The accuracy,precision,recall,F1 value and Kappa coefficient of the model reach 98.24%,98.73%,99.32%,99.02%and 89.77%,respectively,and the Kappa coefficient was increased by 8.52%compared with Unet,indicating that the CBAM+Unet water body recognition model indicated higher recognition accuracy and water edge extraction ability.