CoG-Trans: coupled graph convolutional transformer for multi-label classification of cherry defects.

Item request has been placed! ×
Item request cannot be made. ×
loading   Processing Request
Read Online Read More Add to Saved list
  • Additional Information
    • Subject Terms:
      SURFACE defects; INFRARED imaging; CHERRIES; SWEET cherry; CLASSIFICATION; INFORMATION resources; IMAGE fusion
    • Abstract:
      Diverse categories of defects on the surface of the cherries have different influences on cherries' quality, so simultaneous detection of these defects is essential for their grading. It is a difficult undertaking that requires to investigate the intrinsic category dependencies while taking the category imbalances into account. We treat cherry defect recognition as a multi-label classification task and present a novel identification network called Coupled Graph convolutional Transformer (CoG-Trans). Utilizing the self-attention mechanism and static co-occurrence patterns via our proposed categorical representation extraction Module, we model the relevance of various categories implicitly and explicitly, respectively. Moreover, we design a VI-Fusion module based on the attention mechanism to fuse the visible and infrared information sources. Additionally, we employ asymmetric-contrastive loss to correct the category imbalance and learn more discriminative features for each label. Our experiments are conducted on the VI-Cherry dataset, which consists of 9492 paired visible and infrared cherry images with six defective categories and one normal category manually annotated. The suggested method yields excellent performance compared to previous work, achieving 99.54% mAP on the VI-Cherry dataset. [ABSTRACT FROM AUTHOR]
    • Abstract:
      Copyright of Neural Computing & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)