Predictive modelling of postharvest changes of some selected climacteric fruits from maturity to senescence

The post-harvest handling of fruits and subsequent consumer choices are significantly influenced by the quality changes experienced by the fruits. This study aim is to develop a predictive model to predict the post-harvest changes in some selected climacteric fruits from maturity to senescence. The proximate composition, vitamin content, and physicochemical characteristics of apples, bananas, and guavas were simultaneously examined and captured in JPEG format. A predictive model was developed to establish a correlation between the quality attributes of the selected fruits and their images, using a Convolutional Neural Network (CNN) algorithm. The predictive model was developed as a mobile application using the Python programming language, and the metrics of accuracy, precision, and recall were used. After resizing, every cropped image was split into three sets: the training, validation, and testing sets were 70%, 20%, and 10% of each other, respectively. The training photos were enhanced using Google Colab's "Adam" optimizer while the model was being trained. For Inference, two applications were made: one for the web and another for mobile devices. The web channel was constructed using Flask, HTML, CSS, and JavaScript. The mobile application was made with the Flutter framework and the Dart programming language. The results of the investigation showed that 65 guava fruits and 61 apple fruits were accurately classified, while from the 79 of banana fruits that were analyzed, 77 were accurately categorized  and 2 were mistakenly classified as guava. Predictive model was able to achieve 100%, 97.5%, and 100% for apple, banana, and guava, respectively. 

Keywords: Climacteric fruits, Convolutional Neural Network, Predictive model, Postharvest change, Python