Abstract:
The process of identifying papaya ripeness in indoor and outdoor markets is difficult for consumers because it takes place under different lighting conditions. Such inappropriate choices lead to reduced quality, consumer dissatisfaction, and increased food waste. Conventional methods rely heavily on human judgment, which can be inconsistent and subjective. Thus, a highly reliable, fast, and automated solution is needed to assist consumers in accurately determining fruit ripeness.
For this purpose, transfer learning of pre-trained convolutional neural network (CNN) models was performed, and a system was developed to identify 4 stages of papaya: unripe, half-ripe, ripe, and overripe in indoor and outdoor markets. For this, a new dataset was prepared under different lighting conditions in indoor and outdoor markets. Data enhancement techniques were also applied to increase robustness. CNN models including InceptionV3, MobileNetV2 and DenseNet121 were trained for this solution to identify the most accurate model. For this, experiments were conducted to increase accuracy by increasing the number of epochs. Using these models, a mobile application has been developed to help consumers identify the ripeness level with high accuracy in real-time indoor and outdoor markets.
In order to identify papaya ripeness levels more accurately indoors and outdoors, the results such as accuracy, precision, recall, and F1-score were taken from the trained models and compared with each other, and the 2 models with the highest accuracy were selected. The InceptionV3 model has a high accuracy of 98.27% to identify the papaya ripeness level in the indoor market, and the DenseNet121 model has a high accuracy of 98.71% to identify the papaya ripeness level in the indoor market.