-针对提高小目标信息提取能力、解决目标遮挡问题，对基于单次多核探测器(SSD)的小目标检测方法进行总结,归纳出三种改进途径：一是通过更改骨干网络加强浅层特征图对细节的提取能力；二是通过引入视觉注意力机制获得更有效的特征信息表达；三是将浅层、深层特征信息进行充分融合，提高特征信息利用率及强化各特征层之间的信息交流。最后,对SSD网络结构的改进和多元特征的利用进行展望，旨在为该领域的研究提供参考。

The camera is an important part of the perception system of the unmanned boat, and the target detection of the images taken by it can better help the unmanned boat to complete tasks such as intelligent obstacle avoidance and intelligent search. However, a large number of small targets such as fishing boats and buoys in the image will reduce the target detection accuracy, so it is necessary to study small targets. SSD (Single Shot Multibox Detector) is different from the existing convolutional neural network, making full use of shallow feature information to detect small targets, and has a lot of room for development in the field of small target detection. However, the small target's own characteristic information is not conducive to detection, and the small target with occlusion is more difficult to detect. Therefore, this paper summarizes the small target detection method based on SSD in order to improve the ability of small target information extraction and solve the problem of target occlusion. The first is to enhance the ability of shallow feature maps to extract details by changing the backbone network; the second is to obtain more effective feature information expression by introducing a visual attention mechanism; the third is to fully integrate the shallow and deep feature information to improve the utilization of feature information. rate and strengthen the exchange of information between the feature layers. At the end of this paper, the improvement of the SSD network structure and the utilization of multiple features are prospected, hoping to provide scholars in this field with ideas that can be used for reference.