Abstract:

Object detection is a kind of detection technology which can find and judge the object in the image through computer vision. Different from large and medium object detection, small objects have inherent defects such as less semantic information and small coverage area, which lead to unsatisfactory detection effect of small objects. Therefore, how to improve the detection effect of small objects is still a big problem in the field of computer vision. It outlines the research results in the field of small object detection in recent years at home and abroad. Firstly, it analyzes the definition and detection difficulties of small object. Secondly, it classifies and summarizes the methods that can effectively improve the detection accuracy of small object, and introduces the application and advantages of various methods. Finally, it forecasts and prospects the development trend of small object detection in the future.

Key words: deep learning, computer vision, convolution neural network, small object detection

摘要：

小目标检测是针对图像中像素占比少的目标，借助计算机视觉在图像中找到并判断该目标所属类别的目标检测技术。与目前应用较为成熟的大尺度、中尺度目标检测不同，小目标自身存在着语义信息少、覆盖面积小等先天不足，导致小目标的检测效果并不理想，因此如何提高小目标的检测效果依然是计算机视觉领域的一大难题。对近年来国内外小目标检测领域研究成果进行了梳理，以小目标检测技术为核心，对关于小目标的定义、检测难点进行分析；将能有效提高小目标检测精度的方法进行分类汇总，并介绍了各种方法的应用与优缺点；最后对未来小目标检测领域发展趋势进行了预测与展望。

关键词: 深度学习, 计算机视觉, 卷积神经网络, 小目标检测