Research On Intelligent Ore Sorting Method

China is rich in mineral resources,and the mined minerals are becoming "fine","miscellaneous" and "difficult to select",etc.With the advance of green mining and large-scale mining,the requirements for efficient and intelligent ore screening process are becoming higher and higher.It is of strategic significance and practical demand to carry out research on intelligent ore sorting equipment and guide the deep integration of current artificial intelligence technology and existing ore sorting technology to realize efficient and accurate intelligent ore screening task.In view of the existing problems such as complex ore sorting process,low efficiency and poor accuracy,this paper proposes intelligent classification of coal gangue based on deep learning to solve the demand of accurate classification and recovery of coal gangue in practical engineering application.The main work of this paper is as follows:(1)Based on dual-energy X-ray detection technology,the peripheral hardware solution of intelligent ore separation equipment is systematically proposed.First of all,based on the X-ray line SCAN F dual energy X-ray detector array and digital X-ray source on the basis of imaging equipment configuration to form a system,feeding module transmission at the same time coal gangue,then separation method by high injection ore fall when high-speed separation,finally through Smart-200-PLC and industrial control the whole hardware peripherals together,Realize automatic control and stable work of peripheral hardware circuit of intelligent sorting equipment.(2)Write client software for current hardware peripherals.The whole software client based on Win10＿x64,Qt 5.12.3 development environment,in the form of multi-threaded concurrent development,and the main thread is divided into 7 sub-threads,6 operation modules.Finally,the visualized,real-time and integrated intelligent ore sorting software client is realized.(3)On the basis of dual-energy X-ray imaging equipment,the line array can double probe correction scheme,at the same time to explore a kind of elimination of the influence of the thickness of the ore mapping method,the value of R complete coal gangue image acquisition,according to the collected texture feature and edge details characteristics of coal gangue,further carries on the contrast enhancement and noise reduction work to reduce the follow-up difficulty of neural network training.(4)The mine gangue particle grade on one line is relatively uniform,usually divided into small particle size(radius of 5 cm or less)and large particle size(radius of 5 cm or higher)two kinds of production line,in order to maximize the sorting accuracy,in view of the large particle size of coal gangue separation task,this thesis puts forward a proposed based on dynamic threshold segmentation and depth sorting algorithm of neural network classifier.The dynamic threshold segmentation algorithm was used to eliminate the influence of adhesion shielding and noise interference factors,and the corresponding image of each coal gangue in the 1280×1152 large graph was cut out.The accuracy of the three neural networks for coal gangue image classification is compared.Then,res NET-50 classifier with the best performance was taken as Baseline and improved by ultra-lightweight attention module to achieve further accuracy improvement,and finally completed the establishment of classification algorithm for large particle size ore of intelligent ore sorting equipment.(5)In view of the problems of cumbersome process,low execution efficiency and slow speed in the application of large particle size classification algorithm for ores with small particle size,this thesis adopts target detection technology route to further optimize the back-end algorithm of ore sorting system and proposes a special sorting algorithm for ores with small particle size.In this thesis,two detectors,Yolo V5-S and Yolox-S,were used for comparative experimental verification.Through the visual training process and the final quantitative evaluation of 8 precision indicators and 2 model complexity indicators,the algorithm and the client are finally combined into a complete,integrated,end-to-end industrial equipment to complete the construction of the detection algorithm of the whole intelligent ore separation equipment.