Coal-Rock Boundary Identification Based On Improved YOLOv3 And Research On Human-Simulating Intelligent Control Of Mining Height

Intelligent and unmanned mining is the key to the construction of intelligent mines.Coal and rock boundary recognition and shearer height control are important links to realize intelligent and unmanned mining.YOLOv3 is a target detection algorithm with both detection accuracy and speed,but it still has shortcomings when applied to coal and rock boundary recognition.Therefore,this paper constructs a criterion for the accuracy of coal-rock boundary detection,and proposes an improved YOLOv3 coal-rock boundary recognition method;establishes a coal-rock boundary elevation information extraction method to obtain the actual coal-rock boundary;and on this basis,develop Research on the mining height control strategy of the shearer based on human-simulating intelligent control.The main contents are as follows:(1)Establish a new target detection accuracy evaluation standard for the typical characteristics of coal and rock boundary penetrating type.Because the coal-rock boundary is a through-type,and the typical target detection object is usually block-shaped,the traditional target detection evaluation index(such as accuracy rate,recall rate,etc.)cannot accurately evaluate the accuracy of coal-rock boundary recognition.Therefore,it is necessary to establish accuracy evaluation standards for coal and rock boundaries.This paper proposes to take the ratio of the total projection length of the a priori box in the x direction and(or)y direction to the projection length of the boundary in each direction as the target detection accuracy evaluation standard.(2)Replace the ordinary convolution in the residual block of the original YOLOv3 model with the depth separable convolution,and use the convolution kernel to reduce the dimensionality,thereby reducing the model training parameters and improving the training speed.Related experimental results show that compared with the original model,the training parameter scale of the improved YOLOv3 is reduced by about 80%,and the test time is reduced by about 5%;the accuracy rate in the x direction and y direction is increased by5.85% and 16.99%,respectively.(3)Establish a method for extracting the elevation information of coal and rock boundaries.The aforementioned coal-rock boundary recognition result is expressed in the form of a priori box containing the coal-rock boundary,which cannot directly guide the shearer height control,and further processing is required.In this paper,the cubic spline interpolation algorithm is used to fit the points in the coal-rock boundary recognition a priori frame,and a close-to-real coal-rock boundary curve is obtained.The experimental results show that the fitting difference between the coal and rock boundary obtained by the fitting method and the real boundary is within 3.7%,which meets the requirements.The obtained fitting curve is converted into image coordinates and world coordinates,and the actual elevation information of coal and rock boundaries can be obtained.(4)On the basis of the above coal and rock boundary recognition,a research on the mining height control strategy of the shearer based on human-like intelligent control is carried out.Research shows that compared with traditional PID control,the speed of the shearer height control strategy based on human-like intelligent control is increased by 36.9%,the time to stable is reduced by 6.856 s,and the overshoot is reduced by 4.4%.