The Experiment Research And Design Of Hyperspectral Imaging System

Hyperspectral imaging(HSI)technology is combined with the advantages of traditional imaging and spectroscopy techniques,it obtain the spatial and spectral information of the tested objects,which the spectral information reflects the internal quality of the object such as composition,structure and the spatial information reflects the external quality such as the shape,color and surface defects of the object,etc.As a new,highly efficient,and non-destructive optical imaging technology,it has played a significant role in the rapid non-destructive testing of fruits.In this paper,a hyperspectral imaging system based on a liquid crystal tunable filter(LCTF)has been proposed,which can quickly obtain hyperspectral images of fruit samples in the visible range.By this system,we can obtain the spatial and spectral information of the image,passing through various preprocessing methods and recognition algorithms,achieving rapid identification and classification of fruits.The main research contents are as follows:1.The basic principle of hyperspectral imaging technology and the common system structure were introduced.The mechanism of interaction between fruits ? vegetables and light was analyzed from both macroscopic and microscopic perspectives.The principle of the telephoto imaging system and zoom system was described and the optical path parameter used by the system was calculated.2.Based on the calculated optical path parameters,we designed the system's mechanical structure independently and established the hyperspectral imaging experiment system.The system was simulated and analyzed from the optical path structure.Finally,the performance of the system was analyzed theoretically,and the method for improving the imaging performance of the system was proposed.3.Spectral image preprocessing methods,feature extraction and classification recognition methods were researched.To solve the noise which caused by the hardware of the hyperspectral experimental system,the spectral image black-and-white correction method and various filtering algorithms for removing noise were introduced.For the problem of large amount of data in hyperspectral images,the feature extraction methods of spectral image based on principal component analysis(PCA)is studied.The method of classification and identification of spectral images is introduced,the main researches contain support vector machine(SVM)and artificial neural network(ANN)algorithm.4.An experimental study on the freshness of bananas and the identification of citrus surface defects was conducted.Based on the principal component analysis(PCA)data processing,we compared the recognition accuracy of the SVM and ANN algorithms and the results show that the recognition accuracy is above 85%;Conducted the defect surface experiment research based on different citrus species,using BP neural network algorithm to establish the defect recognition model,the recognition accuracy rate is about 80%.The results show that the hyperspectral imaging technology is feasible for defect recognition detection of citrus fruits.The above research shows that hyperspectral imaging technology is feasible for the research,judgment and classification of fruit quality.