结合目标提取和深度学习的红外舰船检测

传统显著区域提取红外舰船目标检测算法进行图像处理时虚警率高，而深度学习的红外舰船目标检测方法速度慢。针对这些问题，提出了一种将传统的目标提取与深度学习中分类的思想相结合的红外舰船目标检测算法。首先通过高帽变换（TOP-HAT）和 低帽变换（Bottom-HAT）对图像进行处理，然后通过阈值分割方法和归并算法对图像进行候选区域的提取，再运用深度学习中分类的思想完成对目标船舰的检测。通过测试数据集进行实验并对比分析，结果表明改进后的检测算法平均精确度达到83.69%，较之于传统显著区域提取算法精确度提升了8.09%，较之于Faster-R-CNN算法每百张检测时间缩短了2 s。

Infrared ship detection by combining target extraction and deep learning

To solve the problems that the false alarm rate is high when traditional significant region extraction infrared ship target detection algorithm is used for image processing,and the speed of deep learning infrared ship target detection method is slow,an infrared ship target detection algorithm combining the traditional target extraction with the idea of classification in deep learning is proposed.Firstly,the images are processed by High Hat Transform(TOP-HAT) and Low Hat Transform(Bottom-HAT),then the candidate regions are extracted by threshold segmentation method and subsumption algorithm,and then the target ship detection is completed by applying the idea of classification in deep learning.Experiments and comparative analysis are conducted with the test dataset,and the results show that the improved detection algorithm achieves an average accuracy of 83.69%,which is 8.09% better than that of the traditional significant region extraction algorithm,and the detection time per 100 images is shortened by 2 s compared with that of the Faster-R-CNN algorithm.