Intelligent Fruit Quality Assessment Using CNN Transfer Learning Techniques

Fruits are the ultimate source of fibres, vitamins and minerals. Every one of us depends on it to stay healthy. To classify the fruits on basis of their freshness,the image processing methods have been employed to do the classification and identification of the fruits. Different fruits with different dimensions can be identified. A dataset containing fruits like apples, banana, guava in various angles and lighting are used to support classification into 6 categories like fresh and rotten apple, fresh and rotten banana, fresh and rotten guava. The decision regarding the quality is made by processing dataset and studying its features using neural networks. Different neural network architecture like VGG16(91.2%), Inceptionv3(99.6%), Resnet50(67.9%), DenseNet(40.7%) are used and the one with maximum accuracy is found i.e Inception V3.The classifiers like Random Forrest(RF) Classifier(99.39%), Decision-Tree Classifier(DTC)(95.99 %), Logistic-Regression(99. 56 %) and SVM algorithms(99.73%) are integrated along with the neural networks for further reliable outputs among which SVM took the highest accuracy. With the help of transfer learning, the accuracy increased from 99.6% to 99.73%.