Citrus disease detection using convolution neural network generated features and softmax classifier on hyperspectral image data

Authors: YADAV, PAPPU - University Of Florida; BURKS, THOMAS - University Of Florida; FREDERICK, QUENTIN - University Of Florida; Qin, Jianwei - Tony Qin; Kim, Moon; RITENOUR, MARK - University Of Florida

Submitted to: Frontiers in Plant Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/18/2022
Publication Date: 12/7/2022
Citation: Yadav, P., Burks, T.F., Frederick, Q., Qin, J., Kim, M.S., Ritenour, M.A. 2022. Citrus disease detection using convolution neural network generated features and softmax classifier on hyperspectral image data. Frontiers in Plant Science. 13:1043712. https://doi.org/10.3389/fpls.2022.1043712.
DOI: https://doi.org/10.3389/fpls.2022.1043712

Interpretive Summary: Citrus diseases and peel blemishes can limit marketability of citrus crops and in some cases lead to shipping restrictions into certain regions. Proper and timely identification and control of citrus diseases can assure fruit quality and safety, improve production, and minimize economic losses. This study developed an AI-based hyperspectral imaging and classification method for identification of various diseased peel conditions on citrus fruit. Hyperspectral reflectance images were collected in visible and near-infrared wavelength range from Ruby Red grapefruits with normal and common peel diseases, including canker, greasy spot, insect damage, melanose, scab, and wind scar. A classification accuracy over 98% was achieved using a deep learning algorithm based on convolution neural network on the hyperspectral reflectance image data. The method would benefit citrus industry and regulatory agencies (e.g., FDA and USDA APHIS) in ensuring and enforcing the quality and safety standards for the citrus-related food and beverage products.

Technical Abstract: Citrus diseases and peel blemishes can limit marketability of citrus crops and in some cases lead to shipping restrictions into certain regions. This research was aimed at developing an AI-based hyperspectral imaging and classification approach for detecting various peel conditions on citrus fruit. A hyperspectral imaging system was developed for acquiring reflectance images from citrus samples in the spectral region from 450 to 930 nm (92 spectral bands). Ruby Red grapefruits with cankerous, normal, and other common peel diseases including greasy spot, insect damage, melanose, scab, and wind scar were tested. A novel convolution neural network (CNN) was developed to classify images among eight different peel conditions using features from Principal Component Analysis (PCA) for two, three, four and five bands. PCA-based five bands resulted in 100% accuracy, sensitivity, and specificity each. It was also found that the number of PCA-based bands had significant effect on accuracy and sensitivity but not on specificity with two and three bands only achieving 81.71% (73.81%) accuracy and 72.36% (65.32%) sensitivity, respectively. Over 10 trials of randomly selected five bands, the CNN classified images with an average overall accuracy, sensitivity, and specificity of 98.87%, 98.43% and 99.88%, respectively.