Detection and Classification of ventura-blueberries in five levels of ripeness from images taken during pre-harvest stage using Deep Learning techniques

In this article is presented a strategy to detect and classify blueberries inside an "agro-industrial farm" through the combination of Mask R-CNN and a tailored convolutional neural network. First, a dataset is implemented for the training and validation of the proposed strategy. Second, for the captured images the masks related to the blueberries are generated manually to be labeled correctly considering five different ripeness levels. Third, based on the previous step, two models of deep learning for detecting and classifying according to the blueberries’ ripeness levels are implemented. Finally, the proposed models’ test results are shown having a 90.74% of accuracy at the blueberry classification task.