Research On The Detection Of Rock Art In Helan Mountain Based On Deep Learning

The rock art of Helan Mountain are rare relics of ancient civilization,which are of great reference significance for the study of ancient culture.However,some rock paintings are currently being damaged by natural erosion,differentiation and some man-made damage.Therefore,it is urgent to take necessary measures to protect the existing rock paintings.At present,some researchers apply digital electronic technology to the protection research of existing rock paintings.In this process,detection and identification are an important part of it.The Helan Mountain rock art has factors that increase the difficulty of detection and identification,such as low foreground and background distinction,inconspicuous or even missing feature information.In order to improve the recognition accuracy and detection efficiency of Helan Mountain rock art targets,it is necessary to study the existing object detection algorithms.In recent years,object detection algorithms have made great breakthrough.The popular object detection algorithms can be divided into two categories,one is the R-CNN algorithm based on the candidate region(Fast R-CNN,Faster R-CNN,etc.),and the other is the regression-based YOLO,SSD and other algorithms.In this paper,the object detection algorithm based on deep learning is used to detect and identify the target in the Helan Mountain rock art image,and the classical representative algorithms Faster R-CNN,SSD and YOLOv4 among the two types of object detection algorithms are studied.On the basis of the original four types of Helan Mountain rock art target image datasets in the laboratory,the target images added to the category of human images were made to expand the original datasets.Afterwards,Faster R-CNN and SSD algorithms were used to detect and identify the rock paintings,and the effects of different feature extraction networks on the experimental results were compared and analyzed.Finally,the YOLOv4 algorithm is studied to explore the balance between the detection accuracy and detection speed of the model.The YOLOv4 algorithm and its lightweight model YOLOv4-tiny are used to realize the detection and recognition of the rock paintings.The YOLOv4-tiny algorithm is improved by adding the Convolutional Block Attention Module(CBAM)to improve the detection and recognition accuracy of the entire algorithm.The experimental results show that the FPS of the Helan Mountain rock art detection by the YOLOv4 algorithm without Mosaic data enhancement is 47,and the mean average precision reaches 89.80%,which is 5.23%higher than the mean average precision of the YOLOv4 algorithm with Mosaic data enhancement.YOLOv4-tiny without Mosaic data enhancement is used to detect Helan Mountain rock art.With a mean average precision of 82.60%,its detection speed reaches 164 FPS.Adding CBAM attention mechanism to the YOLOv4-tiny algorithm without Mosaic data enhancement has a mean average precision of 85.99%and a detection speed of 127 FPS.Compared with the YOLOv4 algorithm without Mosaic data enhancement,the mean average precision is reduced by 3.81%,the detection speed is increased by about 2.7 times,and a better comprehensive detection performance is achieved.