Submitted to: Proceedings of SPIE
Publication Type: Proceedings
Publication Acceptance Date: 3/17/2008
Publication Date: 4/15/2008
Citation: Qin, J., Burks, T., Kim, M.S., Chao, K., Ritenour, M.A. 2008. Detecting Citrus Canker using Hyperspectral Reflectance Imaging and PCA-based Image Classification Method. Proceedings of SPIE. DOI:10.117/12.786866. Interpretive Summary: Citrus canker is one of the most deleterious diseases that threaten citrus crop production and its industry. A recently developed, portable hyperspectral imaging system was used to investigate reflectance responses from citrus samples in the wavelength range between 400 nm and 900 nm with 99 spectral bands. Ruby Red grapefruits with normal and various diseased skin conditions including canker, copper burn, greasy spot, wind scar, cake melanose, and specular melanose were tested. Hyperspectral reflectance images were analyzed using principal component analysis (PCA) to discriminate cankerous samples from normal and other diseased samples. The overall accuracy for canker detection was 92.7%. The hyperspectral imaging technique for canker disease detection presented in this paper is useful to food scientists, engineers, regulatory government agencies, and food processing industries.
Technical Abstract: A portable hyperspectral imaging system was developed to measure the reflectance images from citrus samples with normal and various common diseased skin conditions in the wavelength range between 400 nm and 900 nm. PCA was used to reduce the spectral dimension of the 3-D hyperspectral image data and extract useful image features that are valuable for differentiating citrus canker from normal and other diseased skin conditions. The third score images obtained from PCA on the hyperspectral data provided best discrimination between canker and normal as well as other diseased skin conditions, and they also could avoid negative effects of stem-ends and calyxes for the canker detection. The simple thresholding method that was used for converting the score images to the binary images was able to segregate the canker lesions from the background of fruit surface. The classifier developed based on counting the number of white pixels in the binary images could differentiate canker from normal fruit skin and other citrus diseases. The overall classification accuracy for citrus canker detection was 92.7%. Copper burn and greasy spot had similar reflectance properties to canker, and the chances for misclassifying the samples with these two diseases were higher than those of the samples with other skin conditions. The treatment of wax on the fruit surface could generate strong specular reflectance, and it may introduce errors for canker detection. The classification method based on the simple threshold value could be affected by different degrees of canker and other diseases on the fruit skin. Further work will be performed to improve the algorithms of image processing and classification for achieving better accuracies for citrus canker detection.