Research On Abnormal Scene Recognition Algorithm Of Coalbed Methane Well Station

With the rapid deterioration of the contemporary human living environment,people pay more and more attention to environmental protection,and coalbed methane(CBM),as a typical clean energy,has received great attention.The high reserves of coalbed methane in China provide a basis for the development of coalbed methane mining in China.However,due to the inflammable and explosive characteristics of coalbed methane,there are serious potential safety hazards in the process of coalbed methane mining.At the same time,because the CBM fields are mostly distributed in remote mountainous areas in a decentralized manner,and the field environment is complicated,if the traditional manual patrol and monitoring method is adopted,on the one hand,safety can not be guaranteed,and on the other hand,human and material resources will be wasted.Therefore,it is necessary to design a field monitoring system based on wireless technology,and this paper focuses on the research of the abnormal scene recognition algorithm of CBM well station used in the system.Firstly,this paper introduces the target detection algorithm commonly used for video monitoring before the appearance of convolutional neural network.The principle of this kind of algorithm is the foreground segmentation technology,which divides the input image into two parts: foreground objects(objects of interest)and background.The principles of representative algorithms such as frame difference method,background subtraction method and optical flow method are summarized.However,since this kind of algorithm cannot effectively detect specific objects,the false detection rate of the experimental scene in this paper is too high,so this kind algorithm is not used.Secondly,combined with the actual experimental environment of this paper,the target detection technology based on convolutional neural network is adopted.According to its principle,it can be divided into target detection algorithm based on region nomination and target detection algorithm based on end-to-end learning.In addition,the principles and processes of representative algorithms,such as R-CNN algorithm,Fast R-CNN algorithm,Faster R-CNN algorithm,YOLO algorithm and SSD algorithm,are introduced.According to the advantages and disadvantages of each algorithm,combined with the complex CBM monitoring scenario in this paper,the main algorithm is determined as SSD algorithm.Since the main body of the traditional SSD algorithm is mostly the VGG-16 model,which requires a large amount of calculation and uses many parameters,the real-time performance of the algorithm is not ideal.Therefore,in combination with the actual situation,this paper chooses to replace the VGG model with the Mobilenet model,and the BN layer is merged..After the introduction of the Mobilenet model,the Mobilenet-SSD(MSSD)target detection algorithm was constructed,which significantly reduced the number of calculations and the numbers of parameters used in multiplication and addition.Meanwhile,Soft NMS algorithm is used to replace the NMS algorithm used in traditional SSD algorithm.These two modifications optimize the performance of the algorithm.Finally,the effectiveness of the algorithm is verified.In response to the experimental requirements,VOC 2007 and part of the added pictures are used in this paper to construct a data set with pedestrians and vehicles as the main body,and then trains and validates the algorithm on this data set.Experimental results show that the proposed algorithm is effective for the recognition of intruding pedestrians and vehicles in the monitoring field,and can identify the target object at the edge of the image,a small part of the occluded target objects,multiple target objects,and small distant target objects.The improved algorithm has a small reduction in running time and an improvement in recognition accuracy.At the same time,the algorithm is applied to other scenarios,and the algorithm is still effective.In conclusion,the accuracy and real-time performance of the algorithm can meet the needs of remote monitoring.