Research Of Detection Method Based On YOLOv3 For Ancient Architectures

With the rapid development of modern science and technology and the increasingly extensive application of communication network technology,people’s social lives are gradually moving towards the era of information,and artificial intelligence has become a hot research topic in today’s society of information.As one of the main research and development fields of artificial intelligence,computer vision has been widely used in security,military,transportation,medical research and social life services and many other fields.Object detection is another key content in the field of computer vision,which mainly studies how to quickly and accurately identify and detect objects from static images or dynamic video streams.What’s more,the method of object detection is focus on how to accurately locate and identify the target in the origin.The intelligent detection system of ancient architectures based on image technology can be used in three dimension digital reconstruction of intelligent ancient architectures,tourism development of intelligent ancient villages and other fields,which has very significant and far-reaching significance value of engineering scientific theory research and application value of engineering reality.However,due to the influence of the style,shape,pattern and texture of ancient architectures,the detection results of current object detectors have the problems of low detection accuracy and inaccurate positioning.In view of the above problems,this paper takes YOLOv3 as the framework and combines the ideas of influence space and pruning to study the detection methods of ancient architectures.Its main work includes:1.Aiming at the problems of low detection accuracy and inaccurate positioning in the detection results of object detector in the process of ancient architecture image detection,a new detective method RNN＿K of ancient architectures based on YOLOv3 network model is proposed.The method named RNN＿K combines density clustering idea with distance clustering idea.The method RNN＿K combines the idea of influence space to generate clustering result set for labeled ancient architecture images firstly.Secondly,the optimal k results are selected in the clustering result set.Then,k-means clustering is carried out to obtain the clustering result,which is used as the anchor of YOLOv3 network.Finally,the RNN＿K algorithm is validated by taking the dataset of ancient architectures as the object.2.Aiming at the problem of large YOLOv3 network model and complex structure,based on the pruning idea,the pruning method of YOLOv3 network model is proposed,which is named YS3.In this method,the weights that do not seriously affect the model performance are removed,so that the YOLOv3 model can become small and fast.The method firstly loads YOLOv3 network,then prunes it and archives the network after pruning.Secondly,quantitative compression of YOLOv3 after pruning is carried out by k-means calculation.Thirdly,sparse training and full training are performed on YOLOv3 before pruning and after compression.Finally,the validity of the method is verified by the evaluation indexes such as model volume,parameter number and compression ratio.3.Based on RNN＿K and YS3,a prototype detection system of ancient architectures is designed and implemented with the ancient architectural images as the objects.The prototype system mainly includes image preprocessing,anchor acquisition and object detection.The system can provide further support for the three dimension reconstruction of ancient architectures.