Review of Small Target Detection based on Deep Learning

Abstract

With a large number of applications of target detection in daily life, the performance requirements of target detection are constantly improving. Many challenges about target detection have been put forward constantly, such as imbalanced samples, fewer pixels, and occlusion of the detected target, all of which bring difficulties for the model to correctly identify the target. Small target detection has always been a difficult point and research hotspot. In recent years, many algorithms for small target detection have been proposed, such as data enhancement, feature fusion, attention mechanism, and super-resolution network structure. According to the characteristics of different network structures, the training strategy can be appropriately adjusted and then applied to different environments, which can greatly improve the detection accuracy of small targets. This paper will introduce the data sets and related small target detection algorithms proposed in recent years, and classify, analyze, and compare the corresponding training strategies.