CAM-K: a novel framework for automated estimating pixel area using K-Means algorithm integrated with deep learning based-CAM visualization techniques

Haciefendioglu K., Mostofi F., Togan V., Basaga H. B.

NEURAL COMPUTING & APPLICATIONS, cilt.34, sa.20, ss.17741-17759, 2022 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 34 Sayı: 20
  • Basım Tarihi: 2022
  • Doi Numarası: 10.1007/s00521-022-07428-6
  • Dergi Adı: NEURAL COMPUTING & APPLICATIONS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Applied Science & Technology Source, Biotechnology Research Abstracts, Compendex, Computer & Applied Sciences, Index Islamicus, INSPEC, zbMATH
  • Sayfa Sayıları: ss.17741-17759
  • Anahtar Kelimeler: Class activation maps, Semantic segmentation, K-Means, Area estimation, Deep learning, CLUSTERING-ALGORITHM
  • Karadeniz Teknik Üniversitesi Adresli: Evet

Özet

This study proposed and implemented a novel framework that can automatically generate accurate area estimation of the identified brick-labeled pixels with the pixel-based intersection of union (IoU) technique. This novel framework employs a combination of fully convolutional neural network with class activation map and K-Means algorithm (CAM-K) to classify, visualize and calculate the pixel areas of brick-labeled images. The existing IoU method based on ground truth and estimated bounding boxes is not suitable for the calculation of localized pixel area. Experiment with our CAM-K framework revealed that it can reliably estimate the pixel areas of the detected object in classified images. Compared with the current state of IoU application, the proposed framework can realize specifically just those targeted pixels objects, and therefore, it can offer a far more realistic IoU metric.