基于深度分割模型的小目标检测

小目标检测广泛应用于视频监控等各种任务,在各领域均有着重要作用。由于待测目标尺寸小、特征弱等原因,目前的检测算法对小目标的检测性能仍值得进一步提升。现有基于设计特征的传统方法在复杂背景的应用场景下检测精度低、鲁棒性弱,基于深度学习的检测算法存在数据集难获取、小目标特征难提取等问题。面向解决低信杂比图像中小目标因面积占比小导致的特征提取难的问题,提出了一个深度分割模型用于小目标检测。为进一步提升检测性能、降低漏检率,充分应用多波段图像信息,设计了一个基于深度分割模型的多波段融合小目标检测方法。在仿真数据集上的实验结果表明,该方法有效提高了小目标检测的准确率,为小目标检测的后续研究提供了新的思路。

Small Target Detection Based on Deep Segmentation Model

Small target detection is widely used in video surveillance and other tasks,which plays an important role in many fields.Due to the small size and weak features of the detected target,the performance of the current detection algorithm for small targets is still worth further improvement.Existing traditional methods based on designed features have low detection accuracy and weak robustness in applications with complex backgrounds.Deep learning based detection algorithms also have problems such as difficult to obtain dataset and extract features for small targets and so on.To solve the problem mentioned above for small targets in low signal-to-clutter ratio (SCR) images,a deep segmentation model for small target detection is proposed.Furthermore,in order to improve the detection performance and reduce the missed detection rate,multi-band image information is made full use of,and a multi-band fusion small target detection method based on the deep segmentation model is designed.Extensive experimental results on the simulation dataset show that the proposed methods effectively improve the accuracy of small target detection and provide novel insights for future researches on small target detection.