Research On Coal Mine Drill Pipe Counting Method Based On Improved RFB And Siamese Network

The realization of automatic counting of coal mine drill pipes through computer vision technology can not only protect the life safety of miners,but also create huge economic value,which has important research significance.At present,the existing drill pipe counting methods mainly include the target detection algorithm represented by the RFB algorithm and the target tracking algorithm represented by the Siamese network algorithm.Due to the serious noise interference of the underground monitoring video image and the lack of effective information,the existing methods have poor accuracy and low efficiency in the detection and tracking of underground targets.For this reason,this paper proposes a method of drill pipe counting combining improved RFB and Siamese network.The main work is as follows:Aiming at the problem of high mismatch rate of feature points of SIIFT algorithm in single scene downhole target detection caused by noise interference and image blur,a target detection algorithm combining linear gray enhancement,Prewitt edge detection and SIFT algorithm is proposed.First,linear grayscale enhancement is used to adjust the grayscale range of the image to the target area.Then,combined with Prewitt edge detection to highlight the edge features of the target image.Finally,Gaussian filtering is used to smooth the noise,and target detection is achieved by matching the edge feature points of the image.The experimental results show that the algorithm can effectively avoid the noise interference of downhole images;in view of the single color and lack of information,the RFB algorithm has low accuracy in multi-scene underground target detection.A target that combines RFB_c receptive field module and high-level superimposed features is proposed.Detection algorithm.First,use the RFB_c receptive field module to expand the receptive field range of the network.Then,the low-level features and the high-level features are merged level by level.Finally,the prediction box is generated by non-maximum suppression.Experimental results show that the improved al gorithm has significantly improved the ability to extract depth features in underground monitoring images.Aiming at the problem that the Siamese network algorithm is inefficient in target tracking downhole due to the interference of redundant information,a target tracking algorithm combining downsampling network,Laplacian interframe sharpening and Siamese network algorithm is proposed.First,the downsampling network is used to reduce the proportion of redundant information.Then,the effective information is further enriched by Laplacian sharpening every other frame.Finally,the residual network is used to extract the depth features of the image,and the target tracking is achieved by measuring the similarity of the depth features of the target image and the image to be tracked.Experimental results show that the al gorithm can improve tracking efficiency while ensuring tracking accuracy.Based on the drill rod counting method,a drill rod counting system is designed and implemented.The system combines local weighted regression and hierarchical clustering to convert the tracking trajectory of the drill chuck into a waveform graph,and realizes coal mine drill pipe counting by judging the effective wave crest.The experimental results show that the system can detect the drill chuck in the underground monitoring video and track it stably,realize the accurate counting of coal mine drill pipes,and visually display the real-time drill pipe quantity.