|Table of Contents|

Data augmentation on vehicle detection in aerial images(PDF)

长安大学学报（自然科学版）[ISSN:1006-6977/CN:61-1281/TN]

Issue:
2023年3期
Page:
95-104
Research Field:
交通工程
Publishing date:

Info

Title:
Data augmentation on vehicle detection in aerial images
Author(s):
CHAI Yan-na1;  SONG Huan-sheng2;  ZHU Jing3
(1. Department of Information and Network Management, Chang'an University, Xi'an 710064, Shaanxi, China; 2. School of Information Engineering, Chang'an University, Xi'an 710064, Shaanxi, China; 3. School of Sciences, Chang'an University, Xi'an 710064, Shaanxi, China)
Keywords:
traffic engineering;  deep learning;  data augmentation;  vehicle detection;  aerial image
PACS:
U491
DOI:
10.19721/j.cnki.1671-8879.2023.03.010
Abstract:
In order to increase the training data of vehicle aerial images for the improvement of the accuracy of the vehicle detection model, and the accuracy of vehicle detection, a generic workflow and a new method of data augmentation was proposed. The workflow consists of a Pluralistic Image Completion generator network and a Tiny YOLOv3 detector network, which were used for new training data generation and border labeling respectively, while also considering related performance indicators in a balanced manner. The generative data augmentation method included a training phase and an augmentation phase. The training phase involved separate training of the generator network and the detector network. The augmentation phase used a generative network to generate new samples and a detector to evaluated the feasibility of these samples for augmentation. After evaluation, the feasible samples formed an enhanced training data set for subsequent detector training. The applying data augmentation to improve the detection performance of vehicles in aerial images of drones was focused on, and a generic workflow and a novel generative data augmentation method was proposed, which only requires the vehicle objects to be annotated with a bounding box in the training dataset and does not require any additional supervision. The detector to be trained with a larger number of instances were allowed, especially if the number of training instances was limited, thereby improving vehicle detection performance. At the same time, the method can be integrated with different generators, indicating that it was a generic method in some sense. The results show that when integrated with Pluralistic and DeepFill, the method can improve their average precision by 25.2% and 25.7%, respectively.3 tabs, 7 figs, 27 refs.

References:

[1] ZHOU W Y,YANG G W,HU S M.Jittor-GAN:A fast-training generative adversarial network model zoo based on Jittor[J].Computational Visual Media,2021,7(1):153-157.
[2]LIU K,SHUAI R,MA L.Cells image generation method based on VAE-SGAN[J].Procedia Computer Science,2021,183:589-595.
[3]LIN T Y,MAIRE M,BELONGIE S,et al.Microsoft COCO:Common objects in context[M]//ECCV.Computer Vision-ECCV 2014.Cham:Springer International Publishing,2014:740-755.
[4]EVERINGHAM M,VAN GOOL L,WILLIAMS C K I,et al.The pascal visual object classes(VOC)challenge[J].International Journal of Computer Vision,2010,88(2):303-338.
[5]冯晓硕,沈 樾,王冬琦.基于图像的数据增强方法发展现状综述[J].计算机科学与应用,2021,11(2):370-382.
FENG Xiao-shuo,SHEN Yue,WANG Dong-qi.A survey on the development of image data augmentation[J].Computer Science and Application,2021,11(2):370-382.
[6]OKAFOR E,SMIT R,SCHOMAKER L,et al.Operational data augmentation in classifying single aerial images of animals[C]//IEEE.Proceedings of 2017 IEEE International Conference on Innovations in Intelligent SysTems and Applications(INISTA).New York:IEEE,2017:354-360.
[7]CHEN H T,LIU C H,TSAI W J.Data augmentation for CNN-based people detection in aerial images[C]//IEEE.Proceedings of 2018 IEEE International Conference on Multimedia & Expo Workshops(ICMEW).New York:IEEE,2018:1-6.
[8]XIE Q Z,LUONG M T,HOVY E,et al.Self-training with noisy student improves ImageNet classification[C]//IEEE.Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2020:10684-10695.
[9]CUBUK E D,ZOPH B,SHLENS J,et al.Randaugment:Practical automated data augmentation with a reduced search space[C]//IEEE.Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2020:3008-3017.
[10]LI Y G,HU G S,WANG Y T,et al.Differentiable automatic data augmentation[M]//ECCV.Computer Vision-ECCV 2020.Cham:Springer International Publishing,2020:580-595.
[11]HO D,LIANG E,STOICA I,et al.Population based augmentation:Efficient learning of augmentation policy schedules[J/OL].Computer Vision and Pattern Recognition[2019-05-14].https://arxiv.org/abs/1905.05393
[12]HATAYA R,ZDENEK J,YOSHIZOE K,et al.Faster AutoAugment:Learning augmentation strategies using backpropagation[M]//ECCV.Computer Vision-ECCV 2020.Cham:Springer International Publishing,2020:1-16.
[13]YU Y,GONG Z Q,ZHONG P,et al.Unsupervised representation learning with deep convolutional neural network for remote sensing images[M]//LNCS.Lecture Notes in Computer Science.Cham:Springer International Publishing,2017:97-108.
[14]ZHU J Y,PARK T,ISOLA P,et al.Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//IEEE.Proceedings of 2017 IEEE International Conference on Computer Vision(ICCV).New York:IEEE,2017:2242-2251.
[15]ZHANG H,SINDAGI V,PATEL V M.Image de-raining using a conditional generative adversarial network[J].IEEE Transactions on Circuits and Systems for Video Technology,2020,30(11):3943-3956.
[16]BAO J M,CHEN D,WEN F,et al.CVAE-GAN:Fine-grained image generation through asymmetric training[C]//IEEE.Proceedings of 2017 IEEE International Conference on Computer Vision(ICCV).New York:IEEE,2017:2764-2773.
[17]HE Y,SCHIELE B,FRITZ M.Diverse conditional image generation by stochastic regression with latent drop-out codes[M]//ECCV.Computer Vision - ECCV 2018.Cham:Springer International Publishing,2018:422-437.
[18]LEE D,LIU S F,GU J W,et al.Context-aware synthesis and placement of object instances[C]//ACM.Proceedings of the 32nd International Conference on Neural Information Processing Systems.New York:ACM,2018:10414-10424.
[19]DVORNIK N,MAIRAL J,SCHMID C.On the importance of visual context for data augmentation in scene understanding[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2021,43(6):2014-2028.
[20]DVORNIK N,MAIRAL J,SCHMID C.Modeling visual context is key to augmenting object detection datasets[M]//ECCV.Computer Vision-ECCV 2018.Cham:Springer International Publishing,2018:375-391.
[21]ZHENG C X,CHAM T J,CAI J F.Pluralistic image completion[C]//IEEE.Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2020:1438-1447.
[22]REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:Unified,real-time object detection[C]//IEEE.Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2016:779-788.
[23]REN S Q,HE K M,GIRSHICK R,et al.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.
[24]ZHANG Y,SHEN Y L,ZHANG J.An improved Tiny-YOLOV3 pedestrian detection algorithm[J].Optik,2019,183:17-23.
[25]RAZAKARIVONY S,JURIE F.Vehicle detection in aerial imagery:A small target detection benchmark[J].Journal of Visual Communication and Image Representation,2016,34:187-203.
[26]DENG J,DONG W,SOCHER R,et al.ImageNet:A large-scale hierarchical image database[C]//IEEE.Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2009:248-255.
[27]YU J H,LIN Z,YANG J M,et al.Generative image inpainting with contextual attention[C]//IEEE.Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2018:5505-5514.

Memo

Memo:
-
Last Update: 2023-06-30