Research on small object detection methods based on deep learning

TL;DR: In this article , the authors systematically describe on small object detection methods based on deep learning, and divide them into four categories based on small-object detection optimization methods, such as data augmentation, multi-scale feature fusion, contextual features, and optimized backbone networks, and analyzes the benefits and drawbacks of each method.

Abstract: The efficiency and accuracy of object detection are steadily improving due to the development and widespread application of deep learning. However, small object detection remains a challenge. When employing mainstream object detection algorithms, small objects have low resolution, little feature information, and weak expressiveness, which leads to missed false detection and poor detection accuracy. This paper systematically describes on small object detection methods based on deep learning, divides them into four categories based on small object detection optimization methods, such as data augmentation, multi-scale feature fusion, contextual features, and optimized backbone networks, and analyzes the benefits and drawbacks of each method, and offers a forecast on future research directions.