Research On Dust Detection Algorithm Based On Depth Learning

Nowadays,Socialism with Chinese characteristics has entered a new chapter,and people’s health problems have risen to the national priority development strategy.Health China,occupational safety and health first,among which the occupational health problems mainly caused by coal miners’ pneumoconiosis are very prominent.Coal mine will produce a lot of dust during underground operation,and it is easy to suffer from pneumoconiosis if it is in the dust environment for a long time,so the detection of dust concentration becomes particularly important.When detecting the dust concentration,it is always affected by the installation position of the sensor,and the sensor can only detect the dust concentration in the relative space,with a small detection range,but can’t have a good grasp of the dust concentration detection in the whole space;At the same time,a large amount of underground dust accumulates,and the detection equipment is easily affected.Being in a high concentration environment for a long time will seriously affect the detection accuracy and service life of the sensor.To solve the above-mentioned problems,the traditional dust sensor is replaced by an industrial camera,which has a wide field of vision and can control the dust situation of the whole job site as a whole.Using the camera can get rid of the installation position limit,and at the same time,it can also expand the dust detection range.Based on this research,a video dust detection algorithm based on deep learning is studied,and a dust detection system is built according to the developed algorithm.The designed system can realize the real-time dust detection,and realize the detection according to the set detection categories.The innovations and main contents of this paper are as follows:(1)The traditional dust detection method is improved.Firstly,the dust image data set is made.Secondly,the image algorithm mainly includes image enhancement,image degradation and image restoration.After background removal,the dust processing results are obtained.Lay the foundation for the following.(2)Aiming at the problem of coal mine dust detection,this paper puts forward a dust detection algorithm based on Yolov5 network from two aspects of accuracy and real-time.In this paper,GhostNet network is proposed to replace Yolov5 backbone network,which can reduce the parameters of backbone network and speed up model training.The mechanism of exporting backbone network to the feature layer of the Neck part to exert attention;At the same time,the CSP structure of the Neck part is changed to Ghost CSP,and the detection accuracy and speed of the improved network model are obviously improved in performance.The real-time performance and accuracy of the dust detection algorithm are verified by the test of the algorithm and the application in the actual situation.After the test,the m AP value of the algorithm can reach 92.11%,and the highest FPS can reach 37,which can meet the real-time and accuracy requirements of the scene.(3)In this paper,12,500 underground dust images were collected,and the dust images were artificially divided into three categories.The first kind is high-concentration dust,which generally appears near mining equipment,the second kind is low-concentration dust,mainly manifested as the first kind of dust after treatment,and the third kind is escape dust,mainly manifested as dust blown away by wind.Compared with the actual field test,the dust detection accuracy can reach more than 90%,and the detection speed can reach more than 35 FPS..(4)Using Visual Studio Code integrated development environment,based on PyQt library for dust software design.The software adopts multi-threading technology,which can be connected to multiple cameras for real-time detection of dust.The software can display the dust detection status in real time,allowing the operation status of the intervention equipment to be considered.Figure 47 Table 10 Reference 84...