| With the advent of the era of artificial intelligence,the industrial field has gradually shifted towards the direction of intelligence.In view of the requirements of automation and intelligence for surface polishing and spraying of large and multifaceted complex structural parts,traditional target detection has disadvantages and limitations such as low accuracy,slow speed and subjectivity of small target detection.Compared with the traditional detection technology,the surface detection technology combined with machine vision has more accurate and higher real-time detection results,saving a lot of labor,material resources and time costs.At present,there are many surface defects detection for workpiece surface defects,automobile body spray paint and other single background color,but there is no professional detection algorithm and system for camouflage spray quality.This paper aims at the workpiece surface camouflage spraying,combined with machine vision,to achieve the workpiece surface spraying quality detection system.The main research contents and results are as follows:(1)Construction of workpiece spraying quality detection data setIn this paper,a sample library for quality detection of camo spraying on the surface of the workpiece is constructed.The sample sources are training and testing samples of defects of spraying surface collected by photographing.The data set is made by web crawlers,the acquisition of existing open source data sets,and the self-production of camouflage spraying methods.This article builds the surface of the defect of camouflage coating data sets collected has six types,including spot,scratch,orange peel,accumulate,stain,burr edge,the six kinds of defects are common in the process of spraying paint,can more fully reflect the actual circumstances of the industrial paint,spray applied to practical industrial environment.(2)Set up the hardware environment of the detection systemA workpiece surface quality inspection system was built on NVIDIA embedded microcontroller platform.The hardware environment of the detection system is composed of image acquisition terminal and embedded front end.The image acquisition end is the Raspberry Pi Camera V2 Camera,and the embedded front end is the NVIDIA Jetson Xavier NX embedded board.(3)A workpiece surface quality detection algorithm is proposedIn view of the requirements of detection accuracy and real-time for the quality inspection of workpiece surface spraying,the K-means++ algorithm was firstly used to re-select the size of Anchor frame Anchor according to the characteristics of the data set in this paper.Cluster generated 6-10 categories through clustering,and 8 categories were finally selected through the comparison of clustering results.Then,by comparing the small target detection algorithms,this paper chooses to remove part of the convolution layer based on YOLOv3 network to improve the real-time detection of the model and reduce the layer number of network model.Because of this system is based on embedded front-end equipment,YOLOv3 model network layer number is more,slow training speed,real time cannot meet the demand of industry,so this article on the basis of YOLOv3-tiny was improved,Dense Net module,introduced in the target network,put forward a new of camouflage coating surface quality detection model,in this data set of tests,the accuracy and real-time performance can reach the standard industrial demand.(4)Realize the workpiece surface quality detection system based on visionThe workpiece surface spraying quality inspection system consists of software and hardware systems.Among them,the hardware system consists of image acquisition terminal,embedded front end and so on.Through the visual based workpiece surface quality detection algorithm model,the software design of the detection system,the realization of the visual based workpiece surface quality detection system,as well as human-computer interaction system interface,the final routine system test,detection system results test. |
PDF Full Text Request