Quality prediction of different pineapple (Ananas comosus) varieties during storage using infrared thermal imaging technique

Infrared thermal imaging is a powerful tool used to monitor the quality and safety of various agricultural products. In this study, infrared thermal imaging was used to evaluate the quality of pineapples during storage. Freshly harvested pineapples of different varieties were stored at 5 °C, 10 °C, and 25 °C for 21 days with 360 samples at each storage temperature. The thermal images were segmented to obtain feature selection based on image parameters. The physicochemical properties of pineapples including firmness, pH, total soluble solids, moisture content, and colour measurements for different varieties were also determined using standard reference methods. Significant differences were found between image parameters and the physicochemical properties of pineapples as well as in the interaction between the applied storage treatments. The prediction performance of pineapple quality was developed using partial least squares regression which obtained R 2 values up to 0.94 for all the quality parameters of the pineapple varieties. The results revealed that 10 °C was found to be the most ideal storage temperature for all the physicochemical properties of the fruit. The variation in the image parameters in relation to the different varieties and storage temperatures were successfully discriminated with overall classification accuracies higher than 97% using support vector machines. Therefore, infrared thermal imaging is feasible as a non-destructive tool for monitoring the fruit quality which could enhance the operation and postharvest handling of pineapples under different storage conditions. • Quality prediction of pineapple varieties using thermal imaging. • PLS models were used as the predictive modelling with R 2 higher than 0.94. • Image parameters were extracted based on the shape and pixel value features. • Significant effects were found between temperature, day, and its interactions. • Overall classification accuracy up to 97% using SVM.