Apple Fruit Classification and Damage Detection Using Pre-trained Deep Neural Network as Feature Extractor

Abstract

AbstractThe specific fruit category identification is necessary for the industries, retail stores, and markets. The whole process needs surveillance of a certain number of people, which takes time and requires the most investment in hiring the labor. The industrial intellectuals are looking forward to a technological solution for the above problems. These labor-intensive tasks can be automated by the use of Computer Vision and Machine Learning. This paper proposes a classification model for different apple fruit varieties by support vector machine (SVM) classifier using hierarchical features extracted from the fully connected layer of the Pre-trained deep convolutional neural network. These extracted features are fed to different classifiers like SVM, Random Forest, Linear Regression, and K-nearest Neighbor. The performance metrics of the proposed classification model are assessed in terms of Accuracy, F1-score, Precision, and Recall. The evaluation metrics show that the SVM classifier provides better results than other classifiers. The SVM classifier trained on the features extracted by ResNet 50 Pre-trained deep neural network achieves Accuracy and precision of 99.1%, F1-score of 95.4%, Recall of 98.6%. The Proposed Results are also compared with the related works in the same domain.KeywordsDeep learningConvolutional neural networkResNet-50Support vector machine