Classification of deep image features of lentil varieties with machine learning techniques

Butuner R., Cinar I., Taspinar Y. S., Kursun R., Calp M. H., Koklu M.

European Food Research and Technology, cilt.249, sa.5, ss.1303-1316, 2023 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 249 Sayı: 5
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1007/s00217-023-04214-z
  • Dergi Adı: European Food Research and Technology
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, ABI/INFORM, Agricultural & Environmental Science Database, Aquatic Science & Fisheries Abstracts (ASFA), BIOSIS, Biotechnology Research Abstracts, CAB Abstracts, Chemical Abstracts Core, Compendex, Food Science & Technology Abstracts, Hospitality & Tourism Complete, Hospitality & Tourism Index, Veterinary Science Database
  • Sayfa Sayıları: ss.1303-1316
  • Anahtar Kelimeler: Classification, Deep learning, Lentil, Machine learning
  • Ankara Hacı Bayram Veli Üniversitesi Adresli: Evet

Özet

Today, image classification methods are widely utilized on agricultural products or in agricultural applications. However, many of these methods based on traditional approaches remain unsatisfactory in terms of obtaining effective results. Within this context, this study aimed to classify lentil images by machine learning algorithms, a current and effective method. In line with this purpose, first of all, a camera system was prepared primarily and a dataset was created by recording lentil grains at 225 × 225 resolution via this system. The dataset contains a total of 33,938 data obtained from 3 lentil species as green, yellow, and red. SqueezeNet, InceptionV3, DeepLoc, and VGG16 architectures, among the CNN methods, were used in order to extract features from the recorded images. Lastly, Artificial Neural Network (ANN), Naive Bayes (NB), Random Forest (RF), Adaptive Boosting (AB), and Decision Tree (DT) algorithms were utilized with the aim of creating models for lentil images’ classification. The classification success of the created machine learning models was calculated and the results were analyzed. The highest classification success with the deep features obtained from the SqueezeNet model, 99.80%, was achieved in the ANN algorithm. The results also revealed that grain size and shape features in image classification can yield much more detailed and precise data than can be obtained practically with manual quality assessment.