【摘要】 针对传统水陆分割线提取算法受光照等环境因素干扰算法准确率降低的问题,提出一种改进FCN的水陆分割线提取方法。去除FCN-8s的最后一个池化层并使用空洞卷积替代第五阶段卷积层,融合第二个池化层的输出特征,保留更多网络学习过程中的重要信息,提高模型分割精度,利用分割的结果确定水陆分割线。实验结果表明,该算法能够克服光照等因素的干扰,对水陆分割线提取准确率达到95%以上,优于传统的提取算法,在水资源管理过程中具有极大的应用价值。更多还原

【Abstract】 Aiming at the problem that the accuracy of traditional algorithm reduces because of environmental factors such as illumination,an improved method for extracting split line of water and land from FCN was proposed.By removing the last pooled layer of FCN-8 sand replacing the fifth-stage convolutional layer with hole convolution,while merging the output characteristics of the second pooling layer,more important information in the network learning process was retained and model segmentation accuracy was improved.The results of segmentation were used to determine the water-land spilt line.Experimental results show that the proposed algorithm can overcome the interference of illumination and other factors,and the extraction accuracy of the split line of water and land is over 95%,which is superior to the traditional algorithm and has great application value in the water resources management process.更多还原