基于超像素与多尺度残差U-Net相结合的遥感图像飞机检测方法

遥感图像中存在飞机很小、角度和位置不确定且背景复杂等问题，从遥感图像中检测飞机是一项重要且具有挑战性的任务，因此，提出一种基于超像素与多尺度残差U-Net(Multi-scale Residual U-Net,MSRU-Net)相结合的遥感图像飞机检测方法。首先对遥感图像进行超像素预分割，将位置相邻且像素特征相似的像素点组成若干个超像素，保持图像进一步分割的有效特征；然后构建多尺度残差U-Net，学习其多尺度判别特征。与传统的飞机检测方法相比，该方法用少量的超像素代替大量像素表达图像特征，降低了图像分割的复杂度，再利用MSRU-Net分割遥感超像素图像，有效检测不同尺度的飞机图像。在公共飞机遥感图像数据集上实验，结果表明，该方法能够有效的检测遥感图像不同尺度的飞机图像，检测精确率达到91.2 %。

Remote Sensing Image Aircraft Detection Method by Combining Superpixel and Multi-scale Residual U-NET

It is an important and challenging task to detect aircraft from remote sensing images (RSI) due to the small size of aircrafts in with uncertain angle and position,and complex background.A remote sensing aircraft detection method is proposed based on superpixel and multi-scale U-Net.Firstly,the image is segmented into super-pixels,and the pixels with similar texture,color and brightness are formed into super-pixels to retain the effective information for further image segmentation.Then,a multi-scale residual U-Net model is constructed to learn multi-scale discriminant features.Compared with the traditional U-Net,the proposed method utilizes a small number of superpixels instead of a large number of pixels to express image features,which can reduce the complexity of image segmentation.Finally,the multi-scale U-Net is employed to segment the superpixel images and detect aircraft with different scales effectively.Experimental results on a public aircraft image dataset show that the proposed method can effectively detect aircrafts with different sizes from RSIs,and The detection accuracy rate reaches 91.2 %.