Spectrum-Based Hybrid Deep Learning for Intact Prediction of Postharvest Avocado Ripeness

This research aims to develop a deep learning model that can classify the ripeness of Haas avocados without damaging the fruit. This data for deep learning includes enhanced visible and near-infrared spectral data, hyperspectral images, and images of ripe and unripe avocado fruit. All three data types were used to train the convolutional neural network model as 1-D and 2-D. Then, these feature extractions were applied to the early fusion technique, and the avocado’s ripeness was classified into five classes: firm, breaking, ripe, overripe, and rotten. In addition, the visible and near-infrared spectral data was enhanced using the multiplicative scatter correction and the standard normal variate techniques. The results showed that the hybrid of deep learning models effectively classified avocado ripeness with an accuracy of 94.43%, 90.46%, and 91.02% for training, validation, and testing of the model.