| As intellectualization plays an increasingly important role in modern manufacturing industry,the traditional spectral imagers based on two-dimensional visual image have been underappreciated in the perception of multidimensional information.This paper presents a new type of snapshot spectral imager to detect three-dimensional(3D)scenes,which is able to obtain the distances of objects in 3D space and the spectral information of each point in space simultaneously by only a single measurement.This can meet the need of the multi-dimensional information development in intelligent manufacturing,providing a new imaging method for the applications of the medical localization diagnosis,the digital space remote sensing and the wisdom agriculture.Based on compressive spectral imaging and integral imaging,some key technologies,such as spectrum acquisition,accurate recognition and spatial orientation of the near field targets,were studied.The problem of obtaining four-dimensional information(three-dimensional space information and one-dimensional spectral information)at the same time on a single sensor was solved,and the rapid acquisition and intelligent analysis of multidimensional information was realized.Firstly,in the issue of the target spectrum acquisition,in order to overcome the problem of large amount of data and time-consuming by traditional spectral imagers,the theory of compressive sensing was introduced,and the compressive spectral imaging technology was used to realize the fast acquisition of the spectral information of the detection scenes.Based on the discrete mathematical transfer model of compressive spectral imaging system,two kinds of coded aperture compressi-ve spectral imaging systems were designed and constructed to obtain spectral data cube by a single frame.The problems of pixel matching and low spatial resolution of image quality were analyzed.An optimized multi-frame joint coding mode was put forward in the static scenes where acquisition time is enough,which can effectively reduce data redundancy and improve the spatial resolution of the reconstructed images.The signal noise ratio(SNR)of the optimized system was increased by 2.5 dB.Secondly,in the noised and complex application environment,the acquired spectral data is used to accurately identify the target.A spectral target recognition algorithm based on sparse representation was studied.And a multi-threshold judgment mechanism was used to effectively recognize the targets in complicated background influenced by noise,which has a higher detection rate and robustness in detection algorithm performance evaluation curve.Another new spectral image restoration algorithm based on spectral unmixing technique was proposed to solve the difficulity of the target spectrum acquisition under ambient weather interference.Compared with the recovery methods based on single-band or full-color image,this method can remove the influence of fog more accurately,and improve the recognition of the target through the linear spectral mixing model.In addition,to get the accurate target spectrum in sample training,a new multiaxial statistical method based on spectral angle and a dimension reduction method based on random projection were adopted,which can be superior to the traditional methods in speed.Third,aiming at the targets in the near field detection,the space positioning study was carried out using integral imaging technology.The mathematical imaging model of the microlens array was analyzed in order to collect the spatial information of the scenes in the compressive spectral imaging system.The shifed transformation matrix from 3D space to two-dimensional subimage array was constructed.By means of the reversibility principle of light,the distance information of the target in 3D space can be reconstructed by compression sensing algorithms.This space positioning method enables the detection system without spatial scanning devices.It can obtain the multi-view angle information of the space scene within an exposure time.The reconstruction method reduces the grid-like noise in the traditional reconstruction image and enhances the contrast between the different targets in the reconstructed depth image.Finally,integrated of the above key technologies,a spectrum sampling model of 3D spatial structure was constructed.A multidimensional system sensing function for spatial and spectral information fusion was established.The four-dimensional data of spatial and spectral information was recovered by compressed sensing algorithms.A microlens array based coded aperture snapshot spectral imager was constructed and verified through numerical simulations.In the experiment,only a single frame image on the detector was required to reconstruct the spectral and spatial information of the targets,effectively solving the problems of target identification and localization.The system has the characteristics of fast imaging,no scanning devices,high compression ratio and large SNR.It can capture not only the static scenes but also the dynamic scenes in video rate.The depth resolution is 2mm and the spectral resolution is lOnm.The results can be further improved by means of multi-frame joint coding or color coded aperture. |
PDF Full Text Request