In recent years,large-scale equipment that’s made of diverse conductive and magnetic alloy materials has been widely used in petrochemical,aerospace,railway and other fields.However,when the equipment has been in service for a long time,there will inevitably be various degrees of damage,if some key components of the equipment appear serious damage,it is quite possible to cause serious consequences.Therefore,it’s significant for our industrial and economic development to explore the defect detection technology.Recently,Magneto-optical imaging(MOI)technology has been largely used in the non-destructive testing of ferromagnetic equipment with its characteristics of visualization,high image resolution and strong detection applicability.This thesis is based on a mobile MOI defect nondestructive testing system,due to the huge amount of magneto-optical video data generated by this system,and it faces the problems of insufficient storage space and transmission bandwidth,which put forward the urgent need for fast and high-quality compression processing of magneto-optical video data.To meet this requirement,this thesis studies the compression processing method aimed at magneto-optical video data,and carries out the optimization design for application implementation based on FPGA.The main research contents are as follows:1.Based on the feature analysis of magneto-optical video data,this thesis proposes a defect image detection and compression method based on background redundant features,which solves the data compression processing problem of the MOI defect detection system.Firstly,a magneto-optic defect image detection method based on background difference is proposed,and the background image is updated adaptively in real time by Surendra background update algorithm,which succeeds in eliminating the background redundancy of magneto-optical video data,and the defect image is extracted,which completes the "first compression" of magneto-optical video data.Secondly,the magneto-optic defect image compression method based on convolutional autoencoder is used to design the image compression network and trained aimed at magneto-optical dataset with the goal to reach the best image reconstruction quality,which realized the high-quality intra-frame compression of magneto-optic defect images.2.In order to realize the high-speed processing of magneto-optical video data,this thesis optimizes the design of magneto-optical video data processing based on FPGA from two perspectives from the perspective of cache and speed.In this thesis,the cache optimization is achieved by the combination way of on-chip and off-chip cache,and with the multi-parallelism design of convolution operation to accelerate the calculating process by 128 times.The incremental update method is used to optimize the calculation in magneto-optical defect image detection,which reduces resource consumption and computation delay.3.The magneto-optical video data processing system designed in this thesis is tested based on FPGA platform.Through the experimental verification,the background redundancy elimination rate of the defect image detection method can be up to 91.95%,and the defect image is extracted completely without missing detection.When the compression ratio of the defect image compression network is 8:1,the PSNR of the reconstructed image is up to 39.8737,and the MS-SSIM is up to 0.9886,the image compression performs good.It takes 10.915 ms to detect and compress a magneto-optical image of size 512×512 based on FPGA,and the power consumption is only 6.034 W.The comprehensive performance of the system is better than on CPU or GPU platform,which has nice practical application value. |