Artificial intelligence system for pineapple variety classification and its quality evaluation during storage using infrared thermal imaging

Pineapple is a tropical fruit that is highly relished for its unique aroma and sweet taste. Monitoring of pineapple quality is essential in order to regulate proper postharvest handling and yield production. In the present study, infrared thermal imaging was used to determine the variety classification and quality attributes of pineapples, specifically total soluble solids (TSS), moisture content, pH, colour changes, and firmness based on various storage conditions (storage temperatures and storage days). Three pineapple varieties were used in this study which are MD2, Morris, and Josapine. A total of 1080 fresh pineapples at a ripening stage of Index 2 were used in this study. The samples were stored at three different storage temperatures i.e. in a cold storage room (5 °C), a controlled refrigerator (10 °C), and an air-ventilated laboratory room (25 °C) with a temperature range of ±2 °C and relative humidity of 85 to 90 %. For each variety, 30 samples were randomly selected for data collection at every seven days intervals (Day 0, Day 7, Day 14, and Day 21). Thermal images of pineapples were acquired at three different varieties at various storage conditions. By using first-order kinetics, the R2 values of quality changes of pineapples ranged from 0.893 to 0.992. The results also demonstrated that the samples stored at 10 °C had the longest shelf life in relation to the changes in firmness and moisture content of the fruit. Principal component analysis was used to develop quantitative prediction models and clustering ability of three different varieties of pineapples. The optimal relations among all the image parameters successfully explained the robustness of the partial least squares (PLS) models which demonstrated a good prediction performance of all quality attributes of pineapples with R2 values of up to 0.94. Several machine learning algorithms including linear discriminant analysis, quadratic discriminant analysis, k-nearest neighbour, support vector machine, decision tree, and Naïve Bayes were applied for the classification of pineapple varieties. The results showed that the support vector machine achieved the best performance from the combination of optimal image parameters with the highest classification rate of 100 %. Convolutional neural networks (CNN) were developed to determine the classification of pineapple varieties with the highest accuracy of 99 % via InceptionV3. The precision, recall, and F1-score demonstrate promising results with the values higher than 0.85 for all pineapple varieties. Multimodal data fusion based on three different CNN architectures including ResNet, VGG16, and InceptionV3 was designed for the classification of pineapple varieties with classification rate up to 92 %. Apart from that, a graphical user interface (GUI)-based software for determination of classification accuracy and quality prediction of the fruit is developed. The application of GUI using the CNN approach can also improve the predictive performance of the fruit classification collected in multi-batch image datasets. Hence, it is noted that the feasibility of infrared thermal imaging coupled with artificial intelligence approaches is a promising technique for assessing the variety classification and the quality parameters of pineapples during storage.

Text FK 202278 IR.pdf Download (1MB)

Actions (login required)

View Item