Recognition And Classification System Of Bottles And Cans For Garbage Classification

Garbage classification and recycling can effectively avoid resource waste and environmental pollution,and has gradually become an important means of global resource recycling and sustainable economic development.As an important recyclable resource,bottles and cans are widely concerned because of its high recycling value and large amount.This thesis studies garbage classification and bottle recycling,and based on the current mainstream target recognition algorithm to realize the effective recognition and classification of bottles and cans.The prototype system is implemented by thesis.The main contents of this thesis are as follows:(1)For model training of bottles and cans,construct and annotate the specific data for bottles and cans.The existing open data(such as Image Net)published data can not effectively suitable for model training due to the small number and single category.Therefore,it is necessary to collect and label data with diversified,real scene and high sample size.This thesis use the camera to transmit the garbage pictures in different scenes in time to solve the inefficient problem of manual shooting.Through image clipping,saturation,contrast and color phase transformation,the data can better express image features.This thesis also realizes the semi-automatic data set annotation,which saves the workload of manual annotation and greatly shortens the time.(2)For accuracy and immediacy of bottles and cans classification,construct a network model for bottles and cans recognition and classification and conduct experimental verification.In the actual scene,the stacking of bottles and cans is chaotic,prone to deformation by external forces,and the features are changeable.In addition,they will overlap each other,which brings difficulties to the recognition.This thesis needs a recognition speed faster algorithm as a training network model,this thesis compared and verified the current YOLO v3 and SSD algorithm,experiments show that YOLO v3 algorithm is more suitable for application in the recognition and classification system of bottles and cans.YOLO v3 is insensitive to small targets,YOLO v3 algorithm need to be improved.GIOU is used to replace the traditional IOU,and K-means clustering algorithm is used to optimize the size of anchor frame.After experimental verification,the optimized YOLO v3 algorithm is about 4% higher than the original algorithm,which suitable application requirements in the actual scene.(3)For application scenarios of bottles and cans classification,build a prototype system for bottles and cans recognition.This thesis based on the training data and network model constructed and use integrating Tensor Flow,which is an open source framework for deep learning,an application prototype system for bottle recognition and classification was built.The system supports two recognition modes,one is image recognition,the other is video recognition.The two modes cooperate with each other,which can greatly improve the recognition efficiency and can be applied in daily life.In summary,this thesis based on deep learning algorithm.The research focuses on three aspects: data collection and labeling,network model selection and prototype system construction for bottles and cans recovery and classification.The experimental results prove that this paper can effectively realize the automatic classification method and system of bottles and cans.It provides a complete solution for garbage classification and has a good application prospect and value.