Research On Feature Recognition And Processing Of Terahertz Focal Plane Detector Imaging

Due to the penetrability,safety and broadband characteristics of terahertz waves,there are broad application prospects in non-destructive testing and biomedical fields,and focal plane terahertz imaging technology is one of the important application branches.Terahertz images often face many problems due to the limitations of the light source,detector performance,and the processing power of the back-end circuit in the imaging system.In terms of uneven light spot,detector size and response rate,circuit thermal noise,etc.,it is very difficult to improve from the hardware.At this time,digital image processing can often get twice the result with half the effort.Under the above background,this article is based on a miniaturized focal plane terahertz imaging system with a size of 384×288 array built in the laboratory.After introducing its operation method,the original data is preprocessed,so that it can be transformed into a subjectively distinguishable system.Image,later analyze the characteristics of the terahertz image and process the image by way of feature recognition.Image stitching can stitch a group of terahertz images with overlapping areas into a high-resolution image with a large field of view,which overcomes the limitation of the size of the detector array.The feature point registration is a very critical step.There are many existing feature point algorithms,but they are often only verified in visible light images,and their performance in terahertz images is basically unknown.In view of the above situation,this article first compares the performance of Harris corner,scale invariant feature transform(SIFT),and speeded up robust feature(SURF)in the terahertz image before and after using Prewitt operator sharpening.The final results show that the number of feature points detected by the Harris corner algorithm is 9 times that of SURF,and the operating efficiency is 116 times that of SIFT.At the same time,the European matching results are much better than the other two algorithms.Afterwards,based on the results of Harris algorithm,the transformation matrix is obtained through Random Sampling Consistency(RANSAC)algorithm,and then the terahertz image is stitched by the weighted fusion method.Multiple sets of experimental results show that the image stitching method can quickly,accurately and effectively stitch multiple terahertz images continuously,and at the same time,the overlapped part has a smooth transition.Terahertz images generally have very obvious interference diffraction regions,and traditional filtering methods cannot handle large areas.In response to this situation,this paper proposes a method based on threshold segmentation to extract and process regions with interference features.First,the particle swarm optimization(PSO)algorithm is used to improve the operating efficiency of the Osto threshold segmentation algorithm by113%.In the interference problem,a new local threshold segmentation algorithm is used to solve the problem of uneven illumination,combined with the result of global threshold segmentation to identify the interference fringe area and replace the gray value,and finally use the results of multiple local threshold segmentation to enhance the image;In the diffraction problem,the diffracted bright spot area is found through global threshold segmentation and the gray scale is replaced,and then the result of global threshold segmentation is used for image enhancement.From the processing results of the terahertz image,it can be seen that the contour of the object is clearer,the influence of interference diffraction on the image is significantly reduced,and the contrast is enhanced.