基于SSD算法的实时无人机识别方法研究
|
摘要
随着无人机广泛运用给人们带来便利的同时,也带来了一些新问题,如无人机非法入侵、碰撞行人等引发安全隐患,因此需要建立一套对指定区域无人机目标进行实时、准确地识别和监控系统。针对图像中无人机目标快速检测和识别问题,提出了两种基于SSD(Single Shot MultiBox Detector)算法的实时无人机识别方法。一种方法是基于SSD获取视频流中的无人机位置,然后利用ResNet网络提取无人机的深度特征,得到1000维特征向量,最后采用KNN(K-Nearest Neighbor)算法对提取的特征进行分类,得到最终的无人机识别结果;在收集的无人机测试集中识别准确率达到了79%。另一种方法是直接将SSD检测到的无人机目标图像送入到AlexNet网络中进行Fine-tuning(微调),在无人机测试集中识别准确率达到了83.75%。实验结果表明,两种方法都能实现实时无人机识别,且准确性方面第二种方法优于第一种方法。同时采用Storm框架,保证高吞吐量地处理数据。
The wide use of the Unmanned Aerial Vehicle(UAV)brings convenience to people,it also causes some new questions. For example,the UAV fly into No-fly zone,and people collision accident which results in safety problem. Therefore,a real-time and accurately recognition,monitor and control system of the UAV target inspecified areas are needed. To settle the problem of fast detecting and recognizing the UAV target in the images,two real-time UAV recognition methods based on SSD algorithm are proposed. In one method,Firstly,the location of UVA from a video stream is grabbed using SSD. Secondly,by utilizing ResNet UVA deep feature is extracted,and then 1000 dimension vectors are gotten. At last,KNN algorithm is used to classify the extracted feature vectors,and then the UVA recognition result is gotten. The recognition accuracy has reached 79% in the collected UVA testing set. In another method,the detected UVA by SSD is sent into AlexNet to do fine-tuning,whose accuracy has reached 83.75% in the UVA testing set. Experiments show that the two method both can accomplish real-time UAV recognition,and the second method is better than the first method on the aspect of accuracy. By incorporating the Storm framework,it ensures that data is processed with high throughput.
The wide use of the Unmanned Aerial Vehicle(UAV)brings convenience to people,it also causes some new questions. For example,the UAV fly into No-fly zone,and people collision accident which results in safety problem. Therefore,a real-time and accurately recognition,monitor and control system of the UAV target inspecified areas are needed. To settle the problem of fast detecting and recognizing the UAV target in the images,two real-time UAV recognition methods based on SSD algorithm are proposed. In one method,Firstly,the location of UVA from a video stream is grabbed using SSD. Secondly,by utilizing ResNet UVA deep feature is extracted,and then 1000 dimension vectors are gotten. At last,KNN algorithm is used to classify the extracted feature vectors,and then the UVA recognition result is gotten. The recognition accuracy has reached 79% in the collected UVA testing set. In another method,the detected UVA by SSD is sent into AlexNet to do fine-tuning,whose accuracy has reached 83.75% in the UVA testing set. Experiments show that the two method both can accomplish real-time UAV recognition,and the second method is better than the first method on the aspect of accuracy. By incorporating the Storm framework,it ensures that data is processed with high throughput.
引文
[1]王靖宇,王霰禹,张科等.基于深度神经网络的低空弱小无人机目标检测研究[J].西北工业大学学报,2018,36(2):258-263.
[2]石磊,王亚敏,曹仰杰等.基于深度卷积神经网络的车型识别[J].计算机科学,2018,45(5):280-284.
[3]Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2014:580-587.
[4]He K,Zhang X,Ren S,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[C]//Proc of the European Conference on Computer Vision(EC-CV).Springer,Cham,2014:346-361.
[5]Girshick R.Fast r-cnn[C]//Proc of the IEEE international conference on computer vision(CVPR),2015:1440-1448.
[6]Ren S,He K,Girshick R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2015:91-99.
[7]Redmon J,Divvala S,Girshick R,et al.You only look once:Unified,real-time object detection[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2016:779-788.
[8]Liu W,Anguelov D,Erhan D,et al.SSD:Single shot multiBoxdetector[C]//Proc of the European Conference on Computer Vision(ECCV).Springer,Cham,2016:21-37.
[9]LeCunY,BottouL,BengioY,et al.Gradient-based learning applied to document recognition[C]//Proc of the IEEE,1998,86(11):2278-2324.
[10]Krizhevsky A,Sutskever I,HintonG E.Imagenet classification with deep convolutional neural networks[C]//Proc of the Advances in neural information processing systems(NIPS),2012:1106-1114.
[11]He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2016:770-778.
[12]Szegedy C,Ioffe S,Vanhoucke V,et al.Inception-v4,inception-resnet and the impact of residual connections on learning[C]//Proc of the Thirty-First AAAI Conference on Artificial Intelligence.USA,2017:4278-4284.
[2]石磊,王亚敏,曹仰杰等.基于深度卷积神经网络的车型识别[J].计算机科学,2018,45(5):280-284.
[3]Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2014:580-587.
[4]He K,Zhang X,Ren S,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[C]//Proc of the European Conference on Computer Vision(EC-CV).Springer,Cham,2014:346-361.
[5]Girshick R.Fast r-cnn[C]//Proc of the IEEE international conference on computer vision(CVPR),2015:1440-1448.
[6]Ren S,He K,Girshick R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2015:91-99.
[7]Redmon J,Divvala S,Girshick R,et al.You only look once:Unified,real-time object detection[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2016:779-788.
[8]Liu W,Anguelov D,Erhan D,et al.SSD:Single shot multiBoxdetector[C]//Proc of the European Conference on Computer Vision(ECCV).Springer,Cham,2016:21-37.
[9]LeCunY,BottouL,BengioY,et al.Gradient-based learning applied to document recognition[C]//Proc of the IEEE,1998,86(11):2278-2324.
[10]Krizhevsky A,Sutskever I,HintonG E.Imagenet classification with deep convolutional neural networks[C]//Proc of the Advances in neural information processing systems(NIPS),2012:1106-1114.
[11]He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proc of the IEEE conference on computer vision and pattern recognition(CVPR),2016:770-778.
[12]Szegedy C,Ioffe S,Vanhoucke V,et al.Inception-v4,inception-resnet and the impact of residual connections on learning[C]//Proc of the Thirty-First AAAI Conference on Artificial Intelligence.USA,2017:4278-4284.