Estimating the Ripeness of Hass Avocado Fruit Using Deep Learning with Hyperspectral Imaging

Rapid ripeness assessment of fruit after harvest is important to reduce post-harvest losses by sorting according the duration until they become ready eat. However, there has been little research on non-destructive estimation and ripening speed avocado fruit. Unlike previous methods, which classify into a few categories (e.g., unripe ripe) or indirectly estimate from its firmness, we developed method using hyperspectral imaging coupled with deep learning regression directly Hass A set 44,096 sub-images 551 images was used train, validate, test convolutional neural network (CNN) predict number days ripeness. Training, validation, samples were generated as train spectral–spatial residual We achieved predictions an average error 1.17 per set. series experiments demonstrated that our approach outperformed classification approaches rely dimensionality reduction techniques such principal component analysis. Our results show potential for combining stage fruit, could help fine-tune processing.