基于YOLO-v3模型压缩的卫星图像船只实时检测

常见的目标检测模型由于模型参数量较大，往往难以部署在无人机、卫星等移动嵌入式设备上。为了对船只进行实时监测，将目标检测模型部署在计算能力较弱的设备上，对基于计算机视觉的卫星图像船只目标检测方法进行研究。针对卫星图像中船舰的形状长宽比例特点，采用K-means++聚类算法选取初始的锚点框; 接着对模型进行多尺度训练，将多尺度金字塔图像作为模型训练的输入; 将YOLO-v3目标检测算法的批归一化层的尺度因子作为通道重要性的度量指标，对YOLO-v3模型进行剪枝压缩。实验结果表明，采用的模型剪枝和压缩方法能有效地对模型进行压缩，模型的参数量减少了91.5%，模型检测时间缩短了60%,极大地减少了系统计算性能的开销。当采用的初始锚点框个数为6个时，平均准确率(mAP)达到77.31%，满足了卫星图像船舰实时性检测的需求。

Abstract

Due to the large number of model parameters, common target detection models were often difficult to be deployed on mobile embedded platforms such as unmanned aerial vehicle and satellite. In order to detect ships in real time, and for the purpose of deploying target detection model in weak computing equipment, the ship detection algorithm based on computer vision was researched. According to the feature of ship shape length ratio and width ratio in satellite images, K-means ++ clustering algorithm was used to select the initial candidate anchor boxes. Multi-scale pyramid images were used as the input of model training. The scale factor of the batch normalization layer of the YOLO-v3 target detection algorithm was taken as the measure index of channel importance, and the YOLO-v3 model was pruned and compressed. Experimental results show that model pruning and compression method can effectively compress the model. The number of parameters of the model size is reduced by 91.5% and the time of model detection is shortened by 60% compared with the original model, which greatly reduces the overhead of system computing performance. When the initial number of candidate boxes is 6, the mAP reaches at 77.31%, which meets the requirements of real-time detecting ship in satellite images.

参考文献

[1] 马啸，邵利民，金鑫，等.舰船目标识别技术研究进展［J］. 科技导报，2019，37(24): 65-78.

[2] 安洁玉，丁斌芬.无人机海监测绘技术应用下舰船遥感图像目标检测［J］. 舰船科学技术，2019，41(24): 187-189.

[3] 胡炎，单子力，高峰.基于Faster-RCNN和多分辨率SAR的海上舰船目标检测［J］. 无线电工程，2018，48(2): 96-100.

[4] REN S Q, HE K M, GIRSHICK R B, et al. Faster R-CNN: towards real-time object detection with region proposal networks ［C］//Proceedings of Advances in Neural Information Processing Systems 28. Montreal, Quebec, Canada: NIPS, 2015: 91-99.

[5] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection ［C］//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE, 2016: 779-788.

[6] 马永杰，宋晓凤.基于YOLO和嵌入式系统的车流量检测［J］. 液晶与显示，2019，34(6): 613-618.

[7] 马啸，邵利民，金鑫，等.改进的YOLO模型及其在舰船目标识别中的应用［J］. 电讯技术，2019，59(8): 869-874.

[8] REDMON J, FARHADI A. Yolov3: an incremental improvement ［J］. arXiv: 1804.02767, 2018.

[9] REDMON J, FARHADI A. YOLO9000: better, faster, stronger ［C］//Proceedings of 2017 IEEE Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017: 6517-6525.

[10] 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.

[11] 刘梦伦，赵希梅，魏宾.基于多尺度多特征卷积神经网络的肝硬化识别［J/OL］.计算机仿真［2020-03-08］.http://kns.cnki.net/kcms/detail/11.3724.TP.20191114.1056.062.html.

[12] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition ［C］//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE, 2016: 770-778.

[13] ZHANG P Y, ZHONG Y X, LI X Q. SlimYOLOv3: narrower, faster and better for real-time UAV applications ［C］//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision Workshop. Seoul, Korea (South): IEEE, 2019: 37-45.

[14] IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift ［C］//Proceedings of the 32nd International Conference on International Conference on Machine Learning. Lille, France: JMLR, 2015: 448-456.

[15] LIU Z, LI J G, SHEN Z Q, et al. Learning efficient convolutional networks through network slimming ［C］//Proceedings of 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017: 2755-2763.

[16] Airbus ship detection challenge ［DB/OL］. https://www.kaggle.com/c/airbus-ship-detection.

[17] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition ［C］//Proceedings of the 3rd International Conference on Learning Representations. San Diego, USA: ICLR, 2015: 1-14.

[18] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector ［C］//Proceedings of the 14th European Conference on Computer Vision. Amsterdam, The Netherlands: Springer, 2016: 21-37.

[19] DAI J, LI Y, HE K, et al. R-FCN: object detection via region-based fully convolutional networks ［C］//Proceedings of the 30th Conference on Neural Information Processing Systems. Barcelona, Spain: NIPS, 2016: 379-387.

陈科峻, 张叶. 基于YOLO-v3模型压缩的卫星图像船只实时检测[J]. 液晶与显示, 2020, 35(11): 1168. CHEN Ke-jun, ZHANG Ye. Real-time ship detection in satellite images based on YOLO-v3 model compression[J]. Chinese Journal of Liquid Crystals and Displays, 2020, 35(11): 1168.

基于YOLO-v3模型压缩的卫星图像船只实时检测

关于本站 Cookie 的使用提示

全站搜索

基于YOLO-v3模型压缩的卫星图像船只实时检测

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索