Multi-View-based Apple Maturity Classification using Similarity Network Fusion versus Classical Machine Learning Classifiers

Computer vision technologies have attracted significant interest in agricultural applications in recent years. In fact, at the core of the artificial intelligence, computer vision and machine learning techniques have proved their effectiveness for various tasks, such as the maturity classification, within the framework of precision agriculture. In this context, the main objective of this work is to classify data invariant to appearance changes, using only few number of 3D apple points in order to detect and recognize the maturity level of coloured apples while analyzing only their shapes. We present an automatic apples maturity classification in order to develop a rapid system and accurate classification of different apple maturity stages. Indeed, the detected red and green apples are at two different phenology stages: green apples at 70% of final size and red apples at advanced ripening. To deal with this issue, and given two views obtained from the eastern side of the tree row and the other from the western side, we have investigated various machine learning classifiers; such as K-Nearest Neighbor, Support Vector Machine and Random Forests, while proposing a similarity network fusion in order to solve the problem of classification by constructing networks samples. The experimental results show that the proposed network fusion-based method provides significantly accurate features for recognizing the maturity level of the captured apples. In fact, an accuracy of 93.33% has been recorded on the challenging PFuji-Size dataset while exploring only few shape-based features.