首页|基于滤波技术和GAN的脉冲展宽分幅相机空间分辨性能提升

基于滤波技术和GAN的脉冲展宽分幅相机空间分辨性能提升

扫码查看
为改善硬件提升的局限,将滤波技术和生成对抗网络(GAN)应用于脉冲展宽分幅相机空间分辨性能提升,并采用图像梯度和成像调制度量化提升效果。研究结果显示,高斯-拉普拉斯算子、同态滤波和GAN均能不同程度提升脉冲展宽分幅相机的空间分辨性能,而将同态滤波的图像增强和GAN的去模糊相结合,能更显著地提升分辨效果,并将图像整体平均梯度由初始的1。456提高到3。399;近轴5 lp/mm区域的局部平均梯度由7。638提高至23。901,平均调制度由7。84%提高至24。32%;远轴5 lp/mm区域的局部平均梯度由3。869提高至11。349,平均调制度由6。07%提高至19。60%。研究结论不仅为脉冲展宽分幅相机空间分辨性能的提升提供了参考方法,而且为滤波和去模糊技术融入超快诊断领域提供了新思路。
Spatial Resolution Performance Improvement of Pulse-Dilation Framing Camera Using Filtering Technique and GAN
Objective The pulse-dilation framing camera is a kind of two-dimensional ultra-fast diagnostic equipment with a temporal resolution of better than 10 ps.However,since its spatial resolution performance is relatively poor due to the design of the long drift region and magnetic focusing,it is difficult to acquire higher-sharpness images in inertial confinement fusion experiments.Although a better effect can be yielded by changing the type and increasing the number of magnetic focusing lenses,some factors such as the stronger magnetic field,larger volume,and non-reuse put forward higher requirements for magnetic shielding,portability,and development cost of the equipment.Therefore,the feasibility of improving the spatial resolution performance of the framing camera is discussed by incorporating deblurring and filtering techniques to improve the spatial resolution performance via hardware optimization.Methods To improve the spatial resolution performance of the pulse-dilation framing camera,we first build a model of the pulse-dilation framing camera with double short magnetic focusing.The imaging characteristics are analyzed by simulating the imaging magnetic field distribution,tracing the electronic motion trajectory,and collecting the imaging distribution of regions and points.Then,the filtering technique principle is introduced,and the resolution plate image of the pulse-dilation framing camera is processed by optimizing the filtering operator.Next,according to the basic principle of the deblurring technique,the generative adversarial network(GAN)model is built by designing the generator network,discriminator network,and loss function.Finally,the collected images are processed by coupling the deblurring with filtering techniques,and the spatial resolution performance improvement is quantified by the relevant evaluation indicators.Results and Discussions Firstly,according to the working principle of the pulse-dilation framing camera with magnetic focusing,the characteristics of the reduction imaging and stripes of the resolution plate are analyzed.Secondly,the GAN is designed by the feature pyramid network,position normalization method,and PReLU activation function.When the resolution plate image is processed by the GAN,the imaging quality of 5 lp/mm is improved obviously[Fig.7(b)].The global average gradient(GAG)of the image is increased from 1.456 to 1.777,the paraxial local average gradient(LAG)rises from 7.638 to 11.117,and the average modulation degree(AMD)is from 7.84%to 12.63%.The abaxial LAG and AMD grow from 3.869 to 5.281,and from 6.07%to 9.33%respectively.Thirdly,the Gauss-Laplacian(G-L)operator,homomorphic filtering(HF)technique,and GAN are combined to smooth,enhance and deblur images.When the resolution plate image is processed by the G-L operator and the HF,the GAG of the image respectively increases to 2.203 and 2.886,the paraxial LAG respectively rises to 14.667 and 16.372,and the AMD is respectively improved to 14.57%and 15.27%.Meanwhile,the abaxial LAG respectively increases to 6.981 and 8.315,and the AMD is respectively improved to 12.33%and 13.31%.When the G-L operator and the HF are respectively integrated into the GAN,the GAG respectively increases to 3.044 and 3.399,the paraxial LAG respectively grows to 22.202 and 23.901,and the AMD is respectively improved to 23.01%and 24.32%.The abaxial LAG respectively rises to 9.647 and 11.349,and the AMD is respectively improved to 16.69%and 19.60%(Figs.7 and 8).Conclusions In the pulse-dilation framing camera,due to the axisymmetric and inhomogeneity effect of the magnetic field,the imaging characteristics are rotation and edge distortion.Furthermore,with enlarging off-axis distance,the imaging distortion gradually increases and the spatial resolution is attenuated.To improve spatial resolution performance and avoid limitations of hardware promotion,we apply the filtering technique and GAN to resolution plate images and quantify the enhancing effect using GAG,LAG,and AMD.The results show that the imaging quality of 5 lp/mm is improved obviously by smoothing,enhancement,and deblurring processes.When the GAN and homomorphic filtering are coupled,the GAG is improved from 1.456 to 3.399,the paraxial LAG is from 7.638 to 23.901,and the AMD is from 7.84%to 24.32%.Additionally,the abaxial LAG is improved from 3.869 to 11.349 and the AMD is from 6.07%to 19.60%.The conclusion provides not only a practical reference method for improving the spatial resolution performance of pulse-dilation framing cameras but also a new idea for the applications of filtering and deblurring techniques in ultra-fast diagnosis.

ultra-fast opticsdiagnostic techniquepulse-dilation framing cameragenerative adversarial networksfiltering techniquespatial resolution performance

徐付涛、白雁力、钟思、宋明诚

展开 >

桂林电子科技大学电子工程与自动化学院,广西 桂林 541004

桂林电子科技大学机电工程学院,广西 桂林 541004

超快光学 诊断技术 脉冲展宽分幅相机 生成对抗网络 滤波技术 空间分辨率性能

国家自然科学基金广西自然科学基金广西自动检测技术与仪器重点实验室主任基金广西图像图形与智能处理重点实验室(桂林电子科技大学)开放基金

118650072022GXNSFAA035561YQ22101GIIP2210

2024

光学学报
中国光学学会 中国科学院上海光学精密机械研究所

光学学报

CSTPCD北大核心
影响因子:1.931
ISSN:0253-2239
年,卷(期):2024.44(11)