Research And Implementation Of Belt Coal Flow Detection Algorithm Based On Machine Vision

In the coal port loading and unloading operation,the belt is the main coal transportation equipment.The position of the current coal flow can be judged by the coal flow information of the specific position on the belt,play an early warning role,but also can prevent the accidental situation such as the fall of the coal flow on the belt.At present,the traditional port belt coal flow detection mainly relies on manual visual inspection,tilt switch,pull rope switch or ultrasonic switch.Manual visual inspection will be limited by light.Due to the particularity of the equipment's working environment,the sensor will often have false detection and missed detection.At the same time,the detection frequency and speed of traditional belt material flow hardware are difficult to meet the actual operation requirements.Improving the accuracy,frequency and speed of detection is an urgent problem to be solved in belt flow detection.Aiming at the limitations of traditional coal flow detection,a belt flow detection method based on machine vision is proposed and applied to coal flow detection to improve the accuracy and speed of coal flow detection.The main research contents are as follows:(1)The coal flow detection algorithm based on machine vision is the core of the research and development of the belt coal flow detection system.The performance of the algorithm directly determines the accuracy and operating speed of the system.The belt flow detection algorithm is improved based on the YOLOv3 network.A lightweight YOLOv3 model method is proposed.On the basis of YOLOv3,a deep separable convolution is introduced to compress the volume of the original model to 1/6 of the original volume,reducing the overall The parameter amount and calculation amount of the model,the model size is reduced from 234 MB to 39 MB.(2)A model compression strategy based on the channel pruning of the BN layer scale factor is proposed.Apply L1 regularization to the scale factor of the BN layer and use it as the scaling factor of the channel pruning to eliminate the unimportant connection relationship in the model,and realizes that the parameter quantity is successfully reduced to 0.085 times of the original model without loss of accuracy,and the model is further reduced The size is reduced to 2.64 MB.The overall compression rate under the two model compression strategies is as high as 98.87%,the speed of the compressed model is increased to 6 times that of the original model,the detection speed is about 89 FPS,and the accuracy rate is as high as 99.80%.Meet the requirements of material flow detection for algorithm accuracy and real-time performance.(3)The belt coal flow detection system based on vision is composed of image and video acquisition and analysis system,data communication system and upper man-machine interface system.According to the real-time video images,the system will send the final detection results,including the information of whether the material flow exists and the current width of the material flow,to PLC and store them in the database.And the system parameters can be modified through the configuration file,which is easy to operate.The belt coal flow detection system was tested in the port of Huanghua,and the accuracy,stability and real-time performance of the system were verified.