多尺度特征DCA融合的海上船舶检测算法研究

doi:10.3778/j.issn.1002-8331.2008-0242

摘要/Abstract

摘要： 为了加强海上交通的安全性，以常见的民用船和军用船为研究对象，针对原始YOLOV3算法在船舶数据集上检测精度不高、目标框出现误检和小目标漏检的问题，提出了改进的船舶检测算法MS-YOLOV3。构建船舶图像数据集Shipdataset，包括数据采集、增强和标签标注，使用维度聚类算法在该数据集中找出合适尺寸的先验框，并应用于相对应的尺度特征图。以Darknet-53的网络框架为基础特征提取网络，增加网络预测尺度，在多尺度特征融合中加入DCA融合策略，提高模型对船舶的检测能力。以MS-YOLOV3为算法框架，采用GIOU作为边框损失函数的参数，提升模型对边界框位置信息的预测准确度。结果MS-YOLOV3与YOLOV3检测算法的对比实验表明，前者在船舶数据集上的精度有7.9个百分点的提升。同时加入的GIOU边框损失，拉低了模型的平均损失，加强了模型的鲁棒性，使得目标框的定位误差大大减小。根据Pascal VOC2007数据集上的训练效果，MS-YOLOV3的平均精度相较于YOLO系列算法、SSD300和Faster-RCNN，精确度优势更加明显。提出的MS-YOLOV3检测模型使得船舶的位置信息和类别精度更加准确。

关键词: 深度学习, 卷积神经网络（CNN）, 船舶检测, 多尺度特征, YOLOV3, 判别相关分析（DCA）, 广义交并比（GIOU）

Abstract: In order to enhance the safety of maritime traffic, an improved ship detection algorithm MS-YOLOV3 is proposed to solve the problems of detection based on YOLOV3 on the ship data set, such as the false detection of target frame and the omission of small targets and low detection accuracy. Firstly, the ship image data set is constructed, including data collection, enhancement and label labeling. The dimension clustering algorithm is used to find the anchor box of appropriate size in the data set and applied to the corresponding scale feature map. Secondly, the model extracts network features based on Darknet-53 and increases network prediction scale. The DCA feature fusion strategy is added to maximize the correlation between high-level features and low-level features, and then the channel splicing is carried out. Measures as said above really improve the detection ability of the model. Finally, GIOU is used as both the parameter of the border optimization function and the border screening index to improve the accuracy of the network in predicting the position information of the border box. The experimental results of MS-YOLOV3 and YOLOV3 show that the accuracy of the former is improved by 7.9 percentage points. At the same time, the added loss of GIOU frame reduces the average loss of the model, strengthens the robustness of the model, and greatly reduces the positioning error of the target frame. According to the training effect on Pascal VOC2007 data set, the average accuracy of MS-YOLOV3 is better than the series algorithm of YOLO, SSD300 and Faster-RCNN. The proposed feature interactive detection model in this paper makes the ship position information and classification more accurate.

Key words: deep learning, convolutional neural network（CNN）, ship detection, multiscale features, YOLOV3, discriminant correlation analysis（DCA）；generalized intersection over union（GIOU）

潘慧, 段先华, 罗斌强. 多尺度特征DCA融合的海上船舶检测算法研究[J]. 计算机工程与应用, 2022, 58(4): 177-185.

PAN Hui, DUAN Xianhua, LUO Binqiang. Research on Marine Ship Detection Based on Multi-scale Feature Fusion and DCA[J]. Computer Engineering and Applications, 2022, 58(4): 177-185.

参考文献

[1] 周知.多维高分辨雷达图像目标识别方法研究[D].成都：电子科技大学，2019.
ZHOU Z.Research on target recognition method of multidimensional high resolution radar image[D].Chengdu：University of Electronic Science and Technology of China，2019.
[2] TRUJILLO V E，HINDERS M K.Container monitoring with infrared catadioptric imaging and automatic intruder detection[J].SN Applied Sciences，2019，1（12）：1-25.
[3] ZHU Z T，PENG S B，XU J，et al.Automatic target recognition method for inverse synthetic aperture sonar imaging[J].Chinese Journal of Acoustics，2018，37（4）：463-476.
[4] 李东洁，李东阁，杨柳.卷积神经网络在动态手势跟踪中的应用[J].计算机科学与探索，2020，14（5）：841-847.
LI D J，LI D G，YANG L.Application of convolutional neural network in dynamic gesture tracking[J].Journal of Frontiers of Computer Science and Technology，2020，14（5）：841-847.
[5] GIRSHICK R，DONAHUE J，DARRELL T.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2004 IEEE Conference on Computer Vision and Pattern Recognition，2004：580-587.
[6] GIRSHICK R.Fast R-CNN[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition，2015：1440-1448.
[7] REN S，HE K，GIRSHICK R.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，39（6）：1137-1149.
[8] LIU W，ANGUELOV D，ERHAN D.SSD：single shot multibox detector[C]//14th European Conference on Computer Vision.Cham：Springer，2016：21-37.
[9] REDMON J，DIVVALA S，GIRSHICK R.You only look once：unified，real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition，2016：779-788.
[10] REDMON J，FARHADI A.YOLO9000：better，faster，stronger[C]//2016 IEEE Conference on Computer Vision and Pattern Recongnition，2016：779-788.
[11] LIN T，DOLLAR P，GIRSHICK R.Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition，2017：936-944.
[12] 刘博，王胜正，赵建森，等.基于Darknet网络和YOLOv3算法的船舶跟踪识别[J].计算机应用，2019，39（6）：1663-1668.
LIU B，WANG S Z，ZHAO J S，et al.Ship tracking recognition based on Darknet-53 and YOLOv3[J].Journal of Computer Application，2019，39（6）：1663-1668.
[13] 魏玮，蒲玮，刘依.改进YOLOv3在航拍目标检测中的应用[J].计算机工程与应用，2020，56（7）：17-23.
WEI W，PU W，LIU Y.Improved YOLOv3 application in aerial target detection[J].Computer Engineering and Applications，202，56（7）：17-23.
[14] REDMON J，FARHADI A.YOLOv3：an incremental improvement[J].arXiv：1804.02767，2018.
[15] LIU M，WANG X，ZHOU A.UAV-YOLO：small object detection on unmanned aerial vehicle perspective[J].Sensors，2020，20（8）：2238.
[16] HE K，ZHANG X，REN S.Deep residul learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition，2016：770-778.
[17] DU M，KE X，ZHOU M.The recognition of maize seeds based on multi-scale feature fusion and extreme learning machine[C]//3rd International Conference on Mechanical Engineering and Intelligent Systems，2015：407-413.
[18] XU C，VASCONCELOS N.Learning receptive fields for pooling from tensors of feature response[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition，2014：835-842.
[19] CHAIB S，LIU H，GU Y.Deep feature fusion for VHR remote sensing scene classification[J].IEEE Transactions on Geoscience and Remote Sensing，2017，55（8）：4775-4784.
[20] BAI Y，HAGHIGHAT M，ABDEL-MOTTALEB M.Kernel discriminant correlation analysis：feature level fusion for nonlinear biometric recognition[C]//24th International Conference on Pattern Recognition，2018：3198-3203.
[21] HAGHIGHAT M，ABDEL-MOTTALEB M，ALHALABI W.Discriminant correlation analysis：real-time feature level fusion for multimodal biometric recognition[J].IEEE Transactions on Information Forensics and Security，2016，11（9）：1984-1996.
[22] YU P，XU D，ZHOU H.Feature level fusion using palmprint and finger geometry based on canonical correlation analysis[C]//2010 3rd International Conference on Advanced Computer Theory and Engineering，Chengdu，2010，5：260-264.
[23] PATEREK A.Improving regularized singular value decomposition for collaborative filtering[C]//Proceedings of KDD Cup and Workshop，2007：5-8.
[24] JIANG B，LUO R，MAO J.Acquisition of localization confidence for accurate object detection[J].arXiv：1807.
11590，2018.
[25] NIE F，HUANG H，CAI X.Efficient and robust feature selection via joint[?]2，1-norms minimization[C]//Advances in Neural Information Processing Systems，2010：1813-1821.
[26] REZATOFIGHI H，TSOI N，GWAK J Y.Generalized intersection over union：a metric and a loss for bounding box regression[C]//2019 IEEE Conference on Computer Vision and Pattern Recognition，2019：658-666.
[27] XU J，MA Y，HE S.3D-GIoU：3D generalized intersection over union for object detection in point cloud[J].Sensors，2019，19（19）：4093.
[28] BOTTOU L.Large-scale machine learning with stochastic gradient decent[C]//19th International Conference on Computational Statistics，2010：177-186.
[29] KHAN S S，AHMAD A.Cluster center initialization algorithm for K-means clustering[J].Pattern Recognition Letters，2004，25（11）：1293-1302.