Apple Variety Recognition Based on Multiview Feature Fusion

Abstract

Nondestructive testing in apple variety identification has become a necessary prerequisite for the industrialization of apple production and increased international competitiveness. According to the characteristics of apple images, this paper designs and implements an apple variety identification algorithm (ACRMV) based on multiview technology. The method comprises two main steps, namely, discriminatory image block selection and a multiview classification algorithm. In the phase of image block selection, local features that occur frequently in one category but seldom in other categories are selected. In the multiview classification stage, a robust multiview classification fusion algorithm is designed based on image block features generated by different descriptors for each view. The experimental results show that ACRMV with the strategy of multifeature fusion and joint training is superior to its corresponding single-view method and to other multiview methods. The discriminative image block selection algorithm uses image blocks with greater discrimination as training data to reduce the influence of redundant data. The proposed method makes full use of the consistency and complementarity among different views to achieve the purpose of merging multiple views and jointly improving recognition performance.