Survey of Small Object Detection Methods Based on Deep Learning

Small object detection has always been a difficult problem in the research area of object detection. In recent years, with the rapid development of deep learning, the research on small object detection has made great progress. Researchers have studied and proposed a series of methods to improve the performance of small object detection from the aspects of network structure, training strategy and data processing. This paper provides a detailed overview of small object detection methods based on deep learning. According to the principle of the methods, the existing small object detection methods were divided into multi-scale prediction, data enhancement technology, feature resolution enhancement, context information, new backbone network and training strategy, a more detailed comparison of these methods was done, and the commonly used small object detection datasets were introduced. Finally, summary and prospect were made combined with the development of the existing technology of small object detection.