基于深度学习算法的高分辨率无人机遥感图像自动分割

Automatic segmentation of high-resolution unmanned aerial vehicle remote sensing images based on deep learning algorithms

高分辨率无人机遥感图像自动分割对于图像的目标识别与检测具有重要意义,为提升图像分割精度,提出基于深度学习算法的高分辨率无人机遥感图像自动分割方法。采用直方图均衡化算法增强遥感图像后,构建基于编/解码器架构的深度学习网络语义分割模型,针对增强后的图像,在编码环节中引入残差模块强化对分割目标有效的特征;在解码环节中,采用多尺度融合模块将低层特征的局部细节信息和高层特征的语义信息相融合。同时针对遥感图像内地物类别不均衡的现象,以带权重的交叉熵为模型损失函数,克服模型选择偏好问题,提升模型分割精度。实验结果显示该方法可准确分割遥感图像内不同类型目标,分割精度达到95%以上。

The automatic segmentation of high-resolution unmanned aerial vehicle remote sensing images is of great significance for target recognition and detection within the image.To improve the accuracy of image segmentation,a deep learning algorithm based automatic segmentation method for high-resolution unmanned aerial vehicle remote sensing images is proposed.After histogram equal-ization algorithm is used to enhance remote sensing image,a deep learning network semantic segmentation model based on codec ar-chitecture is constructed.For the enhanced image,residual module is introduced in the coding phase to enhance the effective features of the segmentation target;In the decoding process,multi-scale fusion module is used to fuse the local details of low-level features with the Semantic information of high-level features.At the same time,in view of the phenomenon that the categories of objects in re-mote sensing images are unbalanced,the weighted cross entropy is used as the model loss function to overcome the problem of model selection preference and improve the model segmentation accuracy.The experimental results show that this method can accurately seg-ment different types of targets in remote sensing images,with a segmentation accuracy of over 95%.