Research On Sorting Method Of Broken Glass Based On Deep Learning

Since the 20 th century,the use of glass has become wider and wider,and more and more cullet produced during the use of glass.A large amount of broken glass is discarded at will,which will not only damage the environment,but also cause economic losses,and now all kinds of resources are facing shortages,which is of great significance for glass recycling.The most important problem to be solved in the recycling and reuse of cullet is color sorting.This paper studies the cullet sorting system based on deep learning,focusing on how to realize the positioning and color recognition of the glass target during the sorting process of cullet.First,the broken glass pieces are uniformly discharged on the conveyor belt by the feeding device,and move forward horizontally along the conveyor belt.After passing through the image collection area,the image is collected by the industrial camera,and then the image is processed by the computer to identify the location and color of the glass.Sort,and finally send the result to the lower computer,and the lower computer controls the corresponding jet valve to sort the corresponding glass jet.In this paper,traditional machine vision algorithms and deep learning target detection algorithms are used to identify the glass positioning and color classification of images.Through the comparison of experimental results,we find the superiority of deep learning in the field of broken glass,in the recognition accuracy and detection The speed is better than traditional machine vision algorithms.The traditional machine vision algorithm first uses the Otsu algorithm to segment the image,and then extracts color features in different color spaces and uses the SVM classifier to achieve color classification.The deep learning algorithm selects currently popular target detection algorithms,including the two-step representative network Faster R-CNN,the one-step representative network YOLOv3,and the latest network YOLOv5.Experiments on the above several different algorithms and comparative analysis based on the results.Aiming at the problem of high missed detection rate of small targets in the above implementation methods,the YOLOv5 network model is improved to make it possible to retain the information of the deep image semantics.Obtain more shallow image features and improve the system’s recognition rate of small targets.And can guarantee the real-time required by the system.The experimental results show that the target detection algorithm based on deep learning is used on the designed multi-color sorting system of broken glass with high recognition accuracy,fast detection speed,wide adaptability and more intelligent.The system can be applied to actual production lines and has good practical value.