改进损失函数的地基云状目标检测算法

doi:10.3778/j.issn.1002-8331.2007-0509

摘要/Abstract

摘要： 云是一种常见的天气现象，云状是天气预测的关键特征。目前，地基云图像中的云状观测主要依赖于气象观测员的目视观测，十分依赖观测员的经验，实时性和效率较低。针对这一问题，提出使用深度学习的方法进行地基云状检测识别。设计了一种新的目标检测边界框损失函数UIoU，将其应用于YOLOv3算法上。并且使用了[K]-means聚类算法重新设计了适用于地基云状数据集的先验框尺寸，使得边界框回归更加精确和稳定。实验结果表明UIoU-YOLOv3相比于原算法精度得到了有效提升，在VOC数据集和地基云状数据集上mAP数值分别提升了3.4个百分点和2.56个百分点。

关键词: 目标检测, 地基云状识别, YOLOv3, DIoU

Abstract: Cloud is a common weather phenomenon, and cloud shape is a key feature of weather prediction. At present, cloud observations in ground nephogram images mainly rely on the visual observations of meteorological observers, and rely heavily on the experience of the observers, with low real-time performance and low efficiency. In response to this problem, the application of deep learning is proposed for ground nephogram detection and recognition. Firstly, a new object detection bounding box loss function UIoU is designed and applied to the YOLOv3 algorithm. And the [K]-means clustering algorithm is used to redesign the anchor box size suitable for ground nephogram datasets, so that the bounding box returns more precisely and stably. Experimental results show that the accuracy of UIoU-YOLOv3 has been effectively improved in comparision with original algorithm, and the mAP values on the VOC datasets and the ground nephogram datasets have been increased by 3.4 percentage points and 2.56 percentage points, respectively.

Key words: object detection, ground nephogram recognition, YOLOv3, distance IoU loss（DIoU）

王胜春, 陈阳. 改进损失函数的地基云状目标检测算法[J]. 计算机工程与应用, 2022, 58(2): 169-175.

WANG Shengchun, CHEN Yang. Ground Nephogram Object Detection Algorithm Based on Improved Loss Function[J]. Computer Engineering and Applications, 2022, 58(2): 169-175.

参考文献

[1] ALIREZA T，FABIO D，CRISTINA C，et al.Neural networks and support vector machine algorithms for automatic cloud classification of whole-sky ground-based images[J].IEEE Geoscience and Remote Sensing Letters，2015，12（3）：666-670.
[2] 鲁高宇，李涛.基于选择性神经网络集成的地基云状识别算法[J].计算机应用与软件，2015，32（3）：185-187.
LU G Y，LI T.Ground nephogram recognition algorithm based on selective neural network ensemble[J].Computer Applications and Software，2015，32（3）：185-187.
[3] 李晨溪，付强，邓芾.基于ESS模型的云状自动识别方法[J].解放军理工大学学报（自然科学版），2016，17（3）：264-269.
LI C X，FU Q，DENG F.Automatic cloud recognition method based on ESS model[J].Journal of PLA University of Science and Technology（Natural Science Edition），2016，17（3）：264-269.
[4] TIAN Z，SHEN C，CHEN H，et al.FCOS：fully convolutional one-stage object detection[C]//Proceedings of the IEEE International Conference on Computer Vision，2019：9627-9636.
[5] 于进勇，丁鹏程，王超.卷积神经网络在目标检测中的应用综述[J].计算机科学，2018，45（S2）：17-26.
YU J Y，DING P C，WANG C.Overview：application of convolution netural network in object detection[J].Computer Science，2018，45（S2）：17-26.
[6] CHEN P，LI Y，ZHOU H，et al.Detection of small ship objects using anchor boxes cluster and feature pyramid network model for SAR imagery[J].Journal of Marine Science and Engineering，2020，8（2）：112.
[7] GIRSHICK R.Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision，2015：1440-1448.
[8] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems，2015：91-99.
[9] HE K，GEORGIA G，DOLLAR P，et al.Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision，2017：2961-2969.
[10] REDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：unified，real time object detection[C]//Proceedings of the IEEE International Conference on Computer Vision，2016：6517-6525.
[11] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//Proceedings of European Conference on Computer Vision，2016：21-37.
[12] REDMON J，FARHADI A.YOLOv3：an incremental improvement[J].arXiv：1804.02767，2018.
[13] YU J，JIANG Y，WANG Z，et al.UnitBox：an advanced object detection network[J].arXiv：1608.01471，2016.
[14] REZATOFIGHI H，TSOI N，GWAK J，et al.Generalized intersection over union：a metric and a loss for bounding box regression[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2019：658-666.
[15] ZHENG Z，WANG P，LIU W，et al.Distance-IoU loss：faster and better learning for bounding box regression[J].arXiv：1911.08287，2019.
[16] LONG J，SHELHAMER E，DARRELL T.Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2015：3431-3440.
[17] HE K，ZHANG X，REN S，et al.Deep residual learning for image recognition[C]//Proceedings of Conference on Computer Vision and Pattern Recognition，2015：770-778.
[18] LIN T，DOLLAR P，GIRSHICK R，et al.Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2017：2117-2125.
[19] 王建仁，马鑫，段刚龙.改进的[K]-means聚类[k]值选择算法[J].计算机工程与应用，2019，55（8）：27-33.
WANG J R，MA X，DUAN G L.Improved [K]-means clustering [k]-value selection algorithm[J].Computer Engineering and Applications，2019，55（8）：27-33.
[20] VINCENT O R，MAKINDE A S，SALAKO O，et al.A self-adaptive [k]-means classifier for business incentive in a fashion design environment[J].Applied Computing and Informatics，2018，14（1）：88-97.
[21] REDMON J，FARHADI A.YOLO9000：better，faster，stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2017：7263-7271.