High-resolution image small target detection method

The invention discloses a high-resolution image small target detection method, which is characterized in that processing is carried out by adopting a partitioning scheme aiming at high resolution, in a training stage, an original image is split into a plurality of sub-images through offline multi-scale cutting, and a network is trained by utilizing the sub-images; in the reasoning stage, prediction is carried out in the form of an online multi-scale sliding window. The method is suitable for detecting high-resolution image targets such as aviation images, satellite remote sensing images and array images. 本发明公开了一种高分辨率图像小目标检测方法，针对高分辨率，采用分块的方案进行处理，在训练阶段，通过离线多尺度裁剪将原始图像拆分成若干张子图，利用子图训练网络；在推理阶段，通过在线多尺度滑窗的形式，进行预测。本专利方法适用于航空图像、卫星遥感图像、阵列式图像等高分辨率图像目标检测问题。