The Optimized Research Of Glass-defect Detection Based On The Distributed System

As the speed of glass inspection increases,the traditional centralized glass-defect inspection system requires higher computer hardware.Due to the poor scalability of the centralized system,the processing speed cannot meet the requirements of the industrial glass-defect detection technology.The batch of image data generated in a short time will cause the system to be broken down,which will affect the entire glass inspection process.Based on the above background,this paper adopts a distributed architecture and designs a threshold segmentation method to complete the online detection of glass-defect images.At the same time,the algorithm is optimized for the distributed computing framework.By adding the data localization module and the pipeline scheduling module,the calculation and storage are localized,the timeliness of data processing is accelerated,and the technical requirements for glass-defect detection are completed.The main work of this paper is as follows:(1)Based on the theoretical analysis of the current glass-defect detection technology,analyze the defects of the centralized system detection and improve the plan of the system,and use a distributed system to complete the design of the glass-defect detection system.(2)Using the distributed computing platform of defect detection for a large number of glass defect images,the data storage structure and detection algorithm of the distributed system were studied,and the distributed detection of online image data was completed using the designed algorithm.(3)Analyze the problems in the MapReduce computing framework,optimize the algorithm for data localization and pipeline scheduling,and complete the performance optimization of the distributed system by changing the system copy-placement strategy and pipeline scheduling strategy.(4)Build Hadoop clusters with different architectures.By comparing the experimental data such as image size,number of nodes,and number of tasks,and finally tested the feasibility and performance of the improved system.The experimental results show that the improved MapReduce computing framework can meet the technical requirements for online detection of glass line defects,and the average processing speed of the traditional distributed framework is increased by 16.8%,with good scalability and scale growth.