YUE Xi,LIANG Yunhao,HE Lei.Research on DT-YOLO Method of Ship Target Detection based on Improved YOLO Algorithm[J].Journal of Chengdu University of Information Technology,2022,37(05):533-537.[doi:10.16836/j.cnki.jcuit.2022.05.008]
基于改进YOLO算法的船舰目标检测DT-YOLO方法研究
- Title:
- Research on DT-YOLO Method of Ship Target Detection based on Improved YOLO Algorithm
- 文章编号:
- 2096-1618(2022)05-0533-05
- Keywords:
- YOLO; ship; target detection; autonomous landing
- 分类号:
- TP391.4
- 文献标志码:
- A
- 摘要:
- 针对舰载无人直升机自主着舰场景理解进场阶段舰船检测问题，提出基于YOLO改进的DT-YOLO舰船目标检测算法。将YOLO的特征提取网络中的残差结构改造为DenseNet中的密集链接结构，并将特征金字塔设计为5层，使得特征图更小，能够进一步提高检测精度。对NMS算法使用线性对数衰减的方式进行增强，有效解决舰船相互遮挡时的漏检情况。最后，建立舰船数据集对算法进行测试和分析。算法平均精准度AP达到94.21%，检测速度到达61.48帧/秒，结果表明算法对舰载无人直升机进场阶段时的舰船检测具有良好的鲁棒性，有效提升了目标较大及相互遮挡时的检测能力。
- Abstract:
- Aiming at the problem of ship detection in the approach stage of understanding the autonomous landing scene of shipborne unmanned helicopters, an improved DT-YOLO ship target detection algorithm based on YOLO is proposed in this paper. Firstly, based on the YOLO algorithm, densely connected modules and transition modules are used to construct a feature extraction network, and five feature pyramids of convolutional layers of different scales are designed to alleviate the problem of inaccurate detection caused by large scale of feature maps. And the NMS algorithm is improved by the linear decay confidence score method, which effectively solves the missed detection when the ships are occluded from each other. Finally, a ship dataset is established and combined with multi-scale training and data augmentation strategies to test and analyze the algorithm. The average accuracy AP of the algorithm reaches 94.21%, and the detection speed reaches 61.48 frames per second. The results show that the algorithm has good robustness for ship detection during the approach phase of the ship-based unmanned helicopter, and effectively improves the detection ability when the target is large and occluded from each other.
参考文献/References:
[1] 王广彪,李华伟,丁文锐,等.无人机在运动舰船上着舰视觉导引技术研究[J].中国电子科学研究院学报,2012(3):275-283.
[2] 刁灿,王英勋,王金提.辅助自动着陆技术[J].系统仿真学报,2008(s1):495-498.
[3] Henriques J F,Caseiro R,Martins P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596.
[4] Danelljan M,Khan F S,Felsberg M,et al.Adaptive color attributes for real-time visual tracking[C].Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,Columbus,USA,2014:1090-1097.
[5] 于洋,李世杰,陈亮,等.基于改进YOLO v2的船舶目标检测方法[J].计算机科学,2019,46(8):332-336.
[6] 苏娟,杨龙,黄华,等.用于SAR图像小目标舰船检测的改进SSD算法[J].系统工程与电子技术,2020,42(5):1026-1034.
[7] 李靓,贺柏根,霍家道.FPN在遥感图像检测中的应用[J].指挥控制与仿真,2019,41(2):130-136.
[8] Redmon J,Divvala S,Girshick R,et al.You only look once:Unified,real-time object detection[C].Proceedings of the IEEE conference on computer vision and pattern recognition,2016:779-788.
[9] 岳晓新,贾君霞,陈喜东,等.改进YOLO V3的道路小目标检测[J].计算机工程与应用,2020,56(21):218-223.
[10] 袁小平,马绪起,刘赛.改进YOLOv3的行人车辆目标检测算法[J].科学技术与工程,2021,21(8):3192-3198.
[11] 张富凯,杨峰,李策.基于改进YOLOv3的快速车辆检测方法[J].计算机工程与应用,2019,55(2):12-20.
[12] Neubeck A,Van Gool L.Efficient non-maximum suppression[C].18th International Conference on Pattern Recognition(ICPR’06).IEEE,2006,3:850-855.
[13] 杨有为,周刚.面向自然场景文本检测的改进NMS算法[J/OL].计算机工程与应用,[2021-05-20].
[14] 杨天乐,钱寅森,武威,等.基于Python爬虫和特征匹配的水稻病害图像智能采集[J].河南农业科学,2020,49(12):159-163.
[15] Dai J,Li Y,He K,et al.R-FCN:Object Detection via Region-based Fully Convolutional Networks[C].Curran Associates Inc.2016.
[16] 吕伏,刘铁.基于学习率自增强的图像识别深度学习算法[J].计算机应用与软件,2021,38(12):268-273.
[17] 杨其利,周炳红,郑伟,等.注意力卷积长短时记忆网络的弱小目标轨迹检测[J].光学精密工程,2020,28(11):2535-2548.
[18] 蔡成涛,吴科君,严勇杰.基于优化YOLO方法机场跑道目标检测[J].指挥信息系统与技术,2018,9(3):37-41.
[19] Bottou L.Large-scale machine learning with stochastic gradient descent[M].Proceedings of COMPSTAT’2010.Physica-Verlag HD,2010:177-186.
[20] 于秀萍,吕淑平,陈志韬.基于YOLO算法的多类目标识别[J].实验室研究与探索,2019,38(3):34-36.
备注/Memo
收稿日期:2022-04-25
基金项目:四川省科技计划重点研发资助项目(2020YFQ0056)