Estimation of citrus canker lesion size using hyperspectral reflectance imaging.

Abstract

The Citrus industry has need for effective approaches to remove fruit with canker before they are shipped to selective international market such as the European Union. This research aims to determine the detectable size limit for cankerous lesions using hyperspectral imaging approaches. Previously developed multispectral algorithms using visible to near-infrared wavelengths, were used to segregate cankerous citrus fruits from other peel conditions (normal, greasy spot, insect damage, melanose, scab and wind scar). However, this previous work did not consider lesion size. A two-band ratio method with a simple threshold based classifier (ratio of reflectance at wavelengths 834 nm and 729 nm), which gave maximum overall classification accuracy of 95.7%, was selected for lesion size estimation in this study. The smallest size of cankerous lesion detected in terms of equivalent diameter was 1.66 mm. The effect of variation of threshold values and number of erosion cycles (applying morphological erosion multiple times to the image) on estimation of smallest detectable lesion was observed. It was found that small threshold values gave better canker classification accuracies, while exhibiting a lower overall classification accuracy. Meanwhile, higher threshold values portrayed the opposite tendency. The threshold value of 1.275 gave the optimum tradeoff between canker classification accuracy, overall classification accuracy and minimal lesion size detection. Increasing the number of erosion cycles reduced detection rates of smaller canker lesions, leading to the conclusion that a single erosion cycle gave the best size estimation results. The erosion kernel of the size 3 mm