PineDL: Deep Learning Based Classification for Maturity Stages of Pineapple through Computer Vision

Maturity is the most essential aspect in determining the storage life and quality of pineapples. The maturity stages of pineapples can be identified by a variety of characteristics, mainly the color of the pineapple. Usually the determination of maturity stages of fruits are done manually, but this results in inaccuracy. A deep learning based Image Processing Method for the classification of pineapples is provided in this paper. The stages of ripening - ripe and unripe is presented in this work. The pineapple images captured undergo preprocessing techniques - binarization, and masking specifically to remove the crown and flower parts from the pineapple. Self collected images undergo augmentation flipping, rotation, zooming and scaling generating 1280 pineapple images. Deep learning models ResNet, VGG16, DenseNet121, EfficientNet, and InceptionV3 are utilized along with 3 optimizers Adam, SGD, and RMSprop for classification. Inceptionv3 provides better result with accuracy of $97.92 \%$ with all optimizers. F1, Precision and Recall curve have been utilized to measure the performance of the model. The misclassifications are represented using the confusion matrix.