Research On Image Measurement Technology Based On Deep Learning

With the advantages of image measurement technology in non-contact,high precision,large dynamic range,rich information,fast measurement speed,high degree of automation,image measurement technology has been more and more widely used.However,in practical application,image measurement technology still has some urgent needs.At present,the main way to improve the measurement accuracy is to upgrade the hardware equipment.However,the hardware replacement and upgrading is a slow process,and often expensive,difficult to meet the general application,and not conducive to large-scale deployment.When considering how to optimize the image measurement,this paper focuses on the edge detection algorithm in the image measurement system.After investigating and analyzing the existing image measurement system and its key technologies,this paper attempts to introduce super-resolution reconstruction technology to achieve the purpose of pixel encryption,so as to make up for the lack of resolution of hardware imaging equipment in disguise,so as to improve the accuracy of image measurement.In this paper,we propose a semantic aware super-resolution reconstruction model based on deep learning.By introducing the common sliding window in the field of object detection,the network can "roughly detect the location of the object" and reduce the amount of computation outside the sliding window,thus greatly improving the reconstruction efficiency.The experimental results show that the proposed optimization model for image measurement task can improve the computational efficiency,and the reconstruction effect can still reach 95.71%of the original VDSR model on PSNR.Compared with the difference based method,the average reconstruction effect is improved by 17.53%on PSNR.When the super-resolution reconstruction algorithm is used to optimize canny edge detection algorithm,the super-resolution reconstruction algorithm is used The edge detection of the resolution reconstruction algorithm is improved by 1.1 1%?1.53%in PSNR,that is,the learning based super-resolution reconstruction algorithm is more suitable for edge detection optimization than the reconstruction based algorithm;finally,the image measurement based on the super-resolution reconstruction algorithm is realized,and the error rate of the measurement results is less than 3%,and the error values are within the size deviation of the workpiece.