Small object detection algorithm based on enhanced anchor and context

• 1、Institute of Network Technology,Beijing University of Posts and Telecommunications

Abstract：Small object detection has important application value in the fields of security, face detection, and automatic driving. Current small object detection has many problems,such as missing detection, and identifingwrong results.The reason for the poor performance is mainly due to the small number of pixels of the small object itself and some design defects of the current target detection algorithm.In view of the shortcomings of the current two-stage method for small object detection, this paper improves the implementation of the anchor and enhances the context information of the small target.At the same time, it adopts a series of methods to compensate for small object detection defects, such as data enhancement for the small object.The experimental results show that small objects can effectively improve the detection performance of small objects by improving the anchor mechanism.At the same time, small objects are sensitive to context information,enhancing the context information of small targets can also effectively improve the performance of small object detection.

Keywords： deep learning object detection small object detection