Miniature Hyperspectral Imaging Technology Based On Linear Filters

The spectral-image cube obtained by hyperspectral imaging can not only characterize the two-dimensional geometric information of the detected space,but also characterize the spectral nature of ground objects at different radiation wavelengths.The advantage of the combination of spectrum and image and the superior spectral resolution ability provide sufficient data information for analyzing the target attributes,so that the objects that are difficult to distinguish in the space image can be detected.And it has been widely used in military,agriculture,medicine and other fields.The hyperspectral imaging system using the linear variable filter（LVF）as the beam splitters is expected to realize a miniaturized spectral imaging system due to its advantages of simple and compact optical structure,which can provide an effective solution for the detection task of portable camouflage targets in the military background.In this paper,a miniaturized hyperspectral imaging system with a LVF as the core is designed and built.In this foundation,a camouflage target detection model based on spatial information is proposed by using the hyperspectral images obtained by the system in this paper,and good detection results are achieved.The main completed work is as follows:（1）According to the requirements of miniaturized applications,a filter-based spectral imaging system scheme is designed based on hyperspectral imaging theory,and the corresponding optical system parameters are determined through simulation and calculation.In this foundation,the optical path debugging and construction of the miniaturized hyperspectral imaging system based on the LVF is completed,and the collection of complete hyperspectral data is realized.Furthermore,by calibrating the hyperspectral imaging system built in this paper in the laboratory,the spectral response parameters such as spectral resolution and center wavelength offset are obtained,and noise removal and reflectivity correction are completed through data preprocessing.（2）According to the application requirements of hyperspectral target detection,classical hyperspectral target detection algorithms such as spectral angle mapping（SAM）algorithm,adaptive cosine estimator（ACE）algorithm and constrained energy minimization（CEM）algorithm are used to establish the corresponding hyperspectral target detection model on the MATLAB platform.Model performance analysis and comparative experiments are carried out based on the standard data set and the data set collected by the hyperspectral imaging system in this paper.The experimental results show that the detection effect of the CEM algorithm is better,and its AUC value can reach more than 0.98.（3）In order to further improve the efficiency of target detection,the improvement method based on the sparse representation algorithm from two aspects is studied in this paper.On the one hand,aiming at the uncertainty of samples when constructing dictionary in sparse representation algorithm,a dictionary construction method based on CEM is studied.By using the CEM algorithm to pre-detect the target and background,the accuracy of the dictionary is improved.On the other hand,a sparse representation algorithm based on spatial weighting is studied.By weighting the spatial information of the 8-neighborhoods of the target point to be detected,its perception range is expanded,and the spatial information of hyperspectral data is fully utilized.The experimental results show that the AUC value of the algorithm in this paper can reach more than 0.99 on several data sets,which effectively improves the target detection efficiency.In summary,the construction of a filter-based hyperspectral imaging system in the spectral range of 500nm-800 nm and the collection of hyperspectral data are realized in this paper.The improved target detection algorithm effectively improves the target detection efficiency,and its AUC value can reach the highest 0.9945.