Integrating reflectance and fluorescence imaging for apple disorder classification

摘要

Multispectral imaging in reflectance and fluorescence modes combined with neural network analysis was used to classify various types of apple disorder from three apple varieties (Honey Crisp, Red Cort, and Red Delicious). Eighteen images from a combination of filter sets and three different imaging modes (reflectance, visible light induced fluorescence, and UV induced fluorescence) were acquired for each apple sample as a base for pixel-level classification into normal or disorder tissue. Two classification schemes, a 2-class and a multiple class, were developed and tested in this study. In the 2-class scheme, pixels were categorized into normal or disorder tissue, whereas in the multiple class scheme, pixels were categorized into normal, bitter pit, black rot, decay, soft scald, and superficial scald tissues. Results indicate that single variety training under the 2-class scheme yielded highest accuracy with total accuracy of 95, 97, and 100 % for Honey Crisp, Red Cort, or Red Delicious respectively. In the multiple-class scheme, the classification accuracy of Honey Crisp apple for normal, bitter pit, black rot, decay, and soft scald tissue was 94, 93, 97, 97, and 94 % respectively. Through variable selection analysis, in the 2-class scheme, fluorescence models yielded higher total classification accuracy compared to reflection models. For Red Cort and Red Delicious, models with only FUV yield more than 95% classification accuracy, demonstrating a potential of fluorescence to detect superficial scald. Several important wavelengths, including 680, 740, 905 and 940 nm, were identified from the filter combination analysis; The results indicate the potential of this technique to accurately recognize different types of disorder on apple.