Classification of Natural and Carbide-Ripened Cavendish Bananas Using Portable Vis-NIR Multispectral Sensor

Abstract

Cavendish banana (Musa acuminata Cavendish) is a climacteric fruit that ripens rapidly after harvest, requiring accurate monitoring to maintain postharvest quality. In Indonesia, calcium carbide is still frequently used to accelerate ripening, although this practice can reduce fruit quality and raise safety concerns. Laboratory-grade spectrophotometers are effective for spectral characterization but are costly and limited to controlled environments, restricting their applicability for practical postharvest evaluation. To address this limitation, this study employed visible and near infrared (Vis-NIR) spectroscopy using the portable and low-cost AS7265X multispectral sensor to classify bananas ripened under different conditions. Spectral data were collected across 410 to 940 nanometers from three treatments, including naturally ripened fruit and fruit ripened with calcium carbide at 32 grams and 96 grams per kilogram of banana, with thirty samples in each group. The spectral irradiance curves showed clear differences, where naturally ripened bananas displayed smoother and lower intensity patterns than carbide-treated fruit. Linear Discriminant Analysis (LDA) was applied to the spectral data and successfully classified all ripening categories with 100% accuracy for both training and test datasets. The model’s performance, evaluated using a confusion matrix, confirmed the perfect classification, indicating that the AS7265X multispectral sensor integrated with LDA offers a reliable, low-cost, and non-destructive approach for assessing banana ripening methods in postharvest quality evaluation.