Research And Implementation Of Plastic Bottle Identification And Positioning System In Sorting Equipment

At present,affected by the economic and technological level,the treatment of plastic waste is mainly direct burial.Such a treatment method will lead to inability to degrade for hundreds of years,waste disposal sites are increasingly depleted,and recyclable resources cannot be utilized,resulting in resource loss.Therefore,research on computer vision-based waste plastic bottle identification and automatic sorting and positioning system is of great significance for large-scale recycling of waste plastics and environmental protection.The research and design of the identification and positioning system of garbage bottles in the sorting equipment proposed in this topic is to research the collection and training of garbage data sets,call models and the positioning of target garbage to achieve the system goals.Therefore,the research content of the topic is mainly divided into the following aspects:1.Analyze the overall structure and functional requirements of the system,and select system hardware(such as Exynos4412 processor,CMOS camera,touch screen,DC-DC converter module,network card module,storage module,communication module)according to the architecture and actual needs.And level matching module,etc.)and system software(such as embedded Linux)to complete the overall design of the waste bottle identification and positioning system in the sorting equipment;2.Implement the embedded software platform on the designed system hardware.It mainly includes the establishment of a deep learning environment platform and embedded cross-compilation environment required for system development,the corresponding analysis,configuration,cutting and transplantation of the two parts of the embedded system(u-boot and Linux kernel),the production and Carry out detailed research on the mounting method,introduce Open CV under the embedded system and detailed source code compilation instructions,and finally generate the image dependency library under the embedded system,and verify it through program migration;3.System data collection and model training.The data set mainly obtains positive and negative samples through its own collection and network,and then annotates the data by manual labeling software and other methods.Set up a training environment platform in a PC,train bottle datasets,and establish related recognition models;4.Research and implementation of system-related algorithms.After having the recognition model,research and analyze the recognition algorithm and positioning algorithm and implement the code on the ARM board;5.Test the system hardware and software platform.First,drive test the designed hardware platform,then simulate the actual preset scene,apply the algorithm transplanted in ARM to identify the bottle,and correct the system parameters and status by testing;Through the above research,the environment platform of the hardware and software of the system was built to realize the identification and positioning of plastic bottles,which proved the feasibility and practicality of the system.The average recognition accuracy rate reaches 90.47%,and the system response time is within 0.5s.According to the comparison,the algorithm YOLOv3 used in this paper is superior to some other algorithms in terms of recognition speed and tediousness of the algorithm,and the improvement of the later system can be carried out more deeply.