Selecting hyperspectral bands and extracting features with a custom shallow convolutional neural network to classify citrus peel defects

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

Submitted to: Smart Agricultural Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/11/2023
Publication Date: 11/13/2024
Citation: Frederick, Q., Burks, T.F., Watson, A., Yadav, P., Qin, J., Kim, M.S., Ritenour, M.A. 2024. Selecting hyperspectral bands and extracting features with a custom shallow convolutional neural network to classify citrus peel defects. Smart Agricultural Technology. 6: Article e100365. https://doi.org/10.1016/j.atech.2023.100365.
DOI: https://doi.org/10.1016/j.atech.2023.100365

Interpretive Summary: Citrus black spot (CBS) is a significant disease that can lead to shipping quarantine and thus cause economic losses for citrus growers. Early detection of CBS would enable mapping and control for the spread of the disease and prevent infected fruits from entering the packing stream. This study developed an AI-based hyperspectral imaging and classification method for detection of citrus fruits infected with CBS. Hyperspectral reflectance images were acquired in visible and near-infrared wavelength range from Valencia oranges with CBS and other common peel diseases. Using features extracted by a convolutional neural network and a support vector machine classifier, a classification accuracy of 94.9% was achieved for differentiating the fruits with CBS and four other peel conditions. The method would benefit the 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 Black Spot (CBS) causes considerable damage to the Florida citrus industry. Early detection of CBS, especially in the presence of other peel blemishes, would enable better mapping and control of CBS spread, reduce wasted fruit, and permit early removal of culls from the packing stream. Oranges whose peels bore the symptoms of four defects/disease (CBS, greasy spot, melanose, and wind scar), as well as a normal control group, were imaged with a hyperspectral imaging system. Principal Component Analysis- (PCA) and Linear Discriminant Analysis (LDA) -based methods were employed to select bands from these images, and a custom convolutional neural network (CNN) for feature extraction was trained with these bands. The extracted features permitted classification of the peel conditions with four classifiers: SoftMax, Support Vector Machines (SVM), Random Forest Classifier (RFC), and K-Nearest Neighbors (KNN). A mean overall accuracy of 94.9% was achieved using an SVM classifier on five bands selected with PCA, and 90.2% with LDA-selected bands. These results show the potential of CNNs to extract features for automated postharvest citrus inspection.