Research On Intelligent Sorting Technology Of Waste Plastic Bottles

With the rapid development of plastics industry in China,people,in addition to enjoying the convenience brought by plastic products,are discarding a lot of plastic bottles,which have seriously damaged the environment.In order to realize the recycling of plastic bottles,it is necessary to classify the waste plastic bottles quickly and accurately.This thesis studies the machine vision-based intelligent technology and system for sorting waste plastic bottles,including the image preprocessing during plastic bottle sorting,method of identifying plastic bottle types and positioning algorithm.After being cleaned and de-labeled,the plastic bottles are evenly dropped into the conveyor through the vibrating hopper and then transmitted to the sorting system.The sorting system mainly involves four parts,which are the lighting system,the image acquisition unit,the image processing unit and the separating unit.First of all,the target bottles are photographed by an industrial camera.Secondly,the images obtained by the image morphology method are preprocessed.Thirdly,the targets are classified and their locations are acquired via the deep learning-based target recognition algorithm,and then the information about types as well as the locations of plastic bottles to be removed is transmitted to the pneumatic separating unit so as to sort the targets.By means of analyzing the actual production demands of different enterprises,the overall structure,work flow and hardware system of intelligent plastic bottle sorting system are designed in this thesis.Meanwhile,the preprocessing is conducted on the collected images to eliminate image distortion and remove color cast.Besides,oriented to the system sorting requirement,the image morphology-based object location method for image processing is studied,and then the deep learning-based object identification algorithm is explored.Moreover,with reference to the forms of plastic bottles,the plastic bottle dataset is established to separately go into the mainstream image classification algorithm and object detection algorithm,and experiment on them respectively.Directing at the problems in the above-mentioned object detection method,the network structure and dataset of Yolo-v3 model are improved accordingly,the feature extraction capability of the model shallow layer network is enhanced and the features of the training data set are integrated.In this way,the rate of identifying small and messy objects is improved,the model generalization ability is strengthened,and the rate of identifying residual tags is effectively enhanced.At the same time,the realtime requirement of the whole system for the detection speed can be ensuredAs observed from the experimental results,this system can identify accurately and quickly,which is of high practical value in sorting the waste plastic bottles under a simulated factory environment.