Multispectral Detection of Citrus Canker Using Hyperspectral Band Selection

The citrus industry has need for effective and efficient approaches to remove fruits with canker before they are shipped to selective international markets. This research was aimed to develop a multispectral method to inspect citrus canker based on band selections of the hyperspectral image data. A total of 960 Ruby Red grapefruits with normal surface, canker, and five common peel diseases including greasy spot, insect damage, melanose, scab, and wind scar were collected during a seven-month harvest period. Hyperspectral reflectance images were acquired in the spectral region of 450 to 930 nm. Correlation analysis (CA) and principal component analysis (PCA) were used for hyperspectral band selections. The canker detection capabilities of the selected bands were evaluated and compared based on the classifications of the pixels in the selected regions of interest (ROIs) of all the peel conditions. A two-band ratio using wavelengths of 729 and 834 nm selected by CA (R834/R729) gave the best overall classification accuracy (95.1%) for the ROI pixel classification, while the highest accuracy using the PCA-selected bands (R907/R718) was 93.1%. CA band selection outperformed PCA in terms of classification performance owing to its supervised nature. The accuracies for three-band and four-band ratios formed by a sequential forward CA selection approach were lower than that of the two-band ratio. Based on the ratio of R834/R729, algorithms for multispectral image processing and classification were developed to differentiate canker from other peel conditions. The overall classification accuracy on a sample basis was 95.7%. The two-band ratio images have great potential to be adopted by a multispectral imaging system for real-time citrus canker detection.