Comparative Analyses of ConvNeXt Architectures for Apple Leaf Disease Classification

Günay Yılmaz A., Özaras Ö. N., Nabiyev V.

6th International Conference on Problems of Cybernetics and Informatics (PCI 2025), Baku, Azerbaycan, 26 - 28 Ağustos 2025, ss.1-9, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/pci66488.2025.11219782
  • Basıldığı Şehir: Baku
  • Basıldığı Ülke: Azerbaycan
  • Sayfa Sayıları: ss.1-9
  • Karadeniz Teknik Üniversitesi Adresli: Evet

Özet

This study investigates the application of voting ensemble methods using ConvNeXt architectures for the detection and classification of apple leaf diseases. The research was conducted in two phases: first, four different ConvNeXt architectures (Tiny, Small, Base, and Large) were trained on the New Plant Disease dataset, and their performances were comparatively analyzed. The optimal model weights from each architecture's training were preserved. In the second phase, four different voting methodologies (hard, soft, maximum, and weighted) were applied to binary combinations of these models, followed by comprehensive performance evaluation. Results demonstrate that among the individual ConvNeXt implementations, the Base architecture achieved superior performance with 99.87 % accuracy in apple leaf disease detection. Notably, the ensemble approach combining Tiny and Base models with either hard or soft voting mechanisms attained perfect classification accuracy (100 %). These findings highlight the potential of ensemble learning techniques integrated with advanced convolutional neural network architectures for developing efficient disease monitoring systems in agricultural applications, particularly for apple cultivation.