计算机科学 ›› 2023, Vol. 50 ›› Issue (6A): 220700006-7.doi: 10.11896/jsjkx.220700006
窦智1, 胡晨光1, 梁竞一1, 郑李明2, 刘国奇1
DOU Zhi1, HU Chenguang1, LIANG Jingyi1, ZHENG Liming2, LIU Guoqi1
摘要: 面向视频的深度学习算法运算复杂度较高,难以满足实时性要求,严重影响了其在边缘计算和实时系统中的应用。轻量化网络成为了研究热点之一,针对大型网络的轻量化网络显著降低了原网络的参数规模,提升了检测速度,但检测精度难以满足工业需求。针对上述问题,文中提出了一种改进的目标检测轻量化网络,在保持小参数规模的前提下,有效提高了检测性能。文中在YOLOv4-tiny骨干网络中添加VIT(Vision Transformer)结构,利用多头自注意力机制使网络可以提取更深层次的物体特征;使用简化后的Bi-FPN,将两检测通道改为三检测通道,增加注意力融合机制,提高模型对图片特征的利用率,提高网络对不同尺寸大小目标的检测精度;使用Ghost卷积替换传统卷积操作,降低网络计算复杂度,减少网络参数。在COCO数据集上进行实验,实验结果表明,在保持网络规模不变的情况下,改进后的算法相比YOLOv4-tiny原网络检测精度取得了明显提升,可同时满足边缘计算及实时系统对深度网络轻量化和准确度的要求。
中图分类号:
[1]BOCHKOVSKIY A,WANG C Y,LIAO H.YOLOv4:OptimalSpeed and Accuracy of Object Detection[J/OL].arXiv:2004.10934,2020. [2]REDMON J,FARHADI A.YOLOv3:An Incremental Improvement[J].arXiv e-prints,2018. [3]REDMON J,FARHADI A.YOLO9000:Better,Faster,Stronger[C]//IEEE Conference on Computer Vision & Pattern Recognition.IEEE,2017:6517-6525. [4]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. [5]LIU W,ANGUELOV D,ERHAN D,et al.Ssd:Single shotmultibox detector[C]//Computer Vision-ECCV 2016:14th European Conference,Amsterdam,The Netherlands.Springer International Publishing,2016:21-37. [6]REN S,HE K,GIRSHICK R,et al.Faster R-CNN:TowardsReal-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(6):1137-1149. [7]HE K,GKIOXARI G,P DOLLÁR,et al.Mask R-CNN][C]//IEEE Transactions on Pattern Analysis & Machine Intelligence.IEEE,2017. [8]LIU S,QI L,QIN H,et al.Path Aggregation Network for Instance Segmentation][C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).IEEE,2018. [9]WANG C Y,BOCHKOVSKIY A,LIAO H Y M.Scaled-yolov4:Scaling cross stage partial network[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2021:13029-13038. [10]WANG B,LE H X,LI W J,et al.Improved mask detection algorithm of YOLO lightweight network[J].Computer Engineering and Applications,2021,57(8):62-69. [11]ZHANG X,ZHANG Y Q,HE B,et al.Research on remotesensing image aircraft target detection technology based on YOLOv4-tiny[J].Optical Technology,2021,47(3):344-351. [12]ZHANG X,WAN T,WU Z,et al.Real-time detector design for small targets based on bi-channel feature fusion mechanism[J].Applied Intelligence,2022,52(3):2775-2784. [13]LU D,MA W Q.Gesture recognition based on improvedYOLOv4-tiny algorithm[J].Journal of Electronics and Information,2021,43(11):3257-3265. [14]TIAN Y,MAO W,YUAN S,et al. A Decision Support System for Power Components Based on Improved YOLOv4-Tiny[J].Scientific Programming,2021,2021:1-11. [15]LIN Y,CAI R,LIN P,et al.A detection approach for bundled log ends using K-median clustering and improved YOLOv4-Tiny network[J].Computers and Electronics in Agriculture,2022,194:106700. [16]WANG L,ZHOU K,CHU A,et al.An improved light-weight traffic sign recognition algorithm based on YOLOv4-tiny[J].IEEE Access,2021,9:124963-124971. [17]HUI T,XU Y L,JARHINBEK R.Detail texture detectionbased on Yolov4-tiny combined with attention mechanism and bicubic interpolation[J].IET Image Processing,2021,15(12):2736-2748. [18]GUO C,LV X,ZHANG Y,et al.Improved YOLOv4-tiny network for real-time electronic component detection[J].Scientific Reports,2021,11(1):22744. [19]DOSOVITSKIY A,BEYER L,KOLESNIKOV A,et al.Animage is worth 16x16 words:Transformers for image recognition at scale[J].arXiv:2010.11929,2020. [20]TAN M,PANG R,LEQ V.EfficientDet:Scalable and Efficient Object Detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).IEEE,2020. [21]HAN K,WANG Y,TIAN Q,et al.GhostNet:More Features From Cheap Operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).IEEE,2020. [22]VASWANI A,SHAZEER N,PARMAR N,et al.Attention isAll You Need[J].arXiv:1706.03762,2017. [23]HINTON G E,SRIVASTAVA N,KRIZHEVSKY A,et al.Improving neural networks by preventing co-adaptation of feature detectors[J].Computer Science,2012,3(4):212-23. [24]HENDRYCKS D,GIMPEL K.Gaussian Error Linear Units(GELUs)[J].arXiv:1606.08415,2016. |
|