Utilizing RGB imaging and machine learning for freshness level determination of green bell pepper (Capsicum annuum L.) throughout its shelf-life

This study investigates the sensory qualities, weight loss and texture changes in green bell peppers as indicator of freshness during storage. It assesses the feasibility of using machine learning methods to monitor freshness changes in the commercial variety ‘ Kyohikari ’ over a 16-d storage period at + 5 °C and 95 % relative humidity (RH). Throughout the storage period, the commercial variety of green bell peppers were stored, and RGB images were captured using a DSLR camera. Sensory assessments and measurements of texture, weight loss, chlorophyll, and carotenoids were conducted at various inteval. Several machine learning approaches- including logistic regression, neural networks, random forests, k-nearest neighbors, and support vector machines, were employed to develop classification and prediction models for fruit freshness during storage intervals of 0, 4, 6, 8, 10, 12, 14, and 16 d. The results indicate that hue angle and chlorophyll content remained unchanged throughout the experiment. However, after 10 d of storage, a 3 % weight loss was observed, accompanied by the detection of off-odors and off-flavors, marking the limit of marketability. The models demonstrated exceptional accuracy in classifying and predicting the freshness of green bell peppers on the storage day, achieving 100 % accuracy with the neural network algorithm. • Green bell peppers from 5 different orchards lost freshness by storage at + 5 ºC. • Machine Learning was used to classify and predict fruit freshness by storage days. • Four methods were applied to build the classification and prediction models. • Neural Network showed 100 % accuracy for classification and prediction models.