多尺度特征融合U-Net的遥感影像黑臭水体智能检测

研究采用卫星遥感技术获取高分辨率遥感影像水体样本数据集，基于深度卷积神经网络从高分辨遥感影像中提取水体并进行黑臭水体智能监测，提出了一种改进U-Net的黑臭水体检测网络模型(IWDNet)。基于U-Net结构引入跳跃式多尺度特征融合，结合通道注意力机制、卷积注意力模块、通道与空间注意力机制生成不同多尺度特征融合注意力机制(MFFAM)模块进行对比，并引入空洞卷积扩大网络感受野，最终实现黑臭水体的识别检测。实验证明：基于跳跃式多尺度融合与CBAM注意力机制的黑臭水体检测网络(MFFCBAM-IWNet)模型有效提升了识别精度，在高分辨遥感影像水体样本数据集上表现最佳，总体精度达98.56%,Kappa系数达0.978 4。

Intelligent Detection of Impaired Water in Remote Sensing Images Based on Multi-scale Feature Fusion U-Net

Satellite remote sensing technology was used to obtain high-resolution remote sensing image water body data set. Based on deep convolutional neural network, water body was extracted from high-resolution remote sensing images and intelligent monitoring of impaired water body was carried out. An Improved Water Detection Network (IWDNet) model based on the proposed U-Net was proposed. Firstly, based on the U-Net structure, skip multi-scale feature fusion was introduced, and different Multi-scale Feature Fusion Attention Mechanisms (MFFAM) modules were generated by combining SE, ECA and CBAM attention mechanisms for comparison. The dilated convolution was introduced to expand the network receptive field. Finally the recognition and detection of impaired water bodies were realized. Experiment results show that the MFFCBAM-IWNet model based on skip multi-scale fusion and CBAM attention mechanism effectively improves the recognition accuracy, and performs best on the high-resolution remote sensing image water body data set. The overall accuracy is 98.56%, and the Kappa coefficient is 0.9784.