A novel geographic origin identification method based on multi-segment recurrence plot coding and residual attention mechanism

Abstract

Listen

Translate article

In this study, a novel geographic origin identification method is proposed, combine with multi-segment Recurrence plot coding and residual attention mechanism to enhance the performance of apple geographic origin discrimination. First, the near-infare spectra of apples are divided into four segments: 729–798 nm, 789–858 nm, 849–918 nm, and 906–975 nm. Second, Recurrence plot coding is adopted to convert the above four segments into images. Third, a novel convolutional neural network composed of residual attention mechanism is designed to discriminate the images. The results show that the proposed method can improve the performance of apple geographic origin identification. The accuracies of training, validation, and test set are 0.994, 0.988, and 0.969, respectively, the precisions of the test set are 0.975, 0.974, 0.951, and 0.975 corresponding to four different origins, the recalls of the test set are 0.975, 0.963, 0.963, and 0.975. Meanwhile, the proposed method can considerably decrease the complexity of computation. This study provides a strategy to enhance the classification performance of near-infare spectra and an intelligent technology to trace the apple geographic origin.