Algerian Journal of Renewable Energy and Sustainable Development
Volume 6, Numéro 2, Pages 194-204
2024-12-15
Semi-supervised Teacher-student Architecture For Weed Classification
Authors : Nassima Bousahba . Mahamed Abdelmadjid Allali . Kawther Karnache . Imane Fettah . Faiza Boudjeltia .
Abstract
Recent advances in deep learning have demonstrated remarkable success in image classification tasks, particularly in distinguishing between weeds and crops. This research focuses on weed classification using the Weed_Ch32 dataset, which comprises 5,023 smartphone-captured images of 32 weed species collected from Algeria's National Plant Protection and Institute, Labyed Medjadja-Chlef. Our evaluation encompasses several established models including VGG16, Xception, Inception, Mobilenet, and Mobilenet v2. In our experiments, Mobilenet v2 achieved 88% accuracy in supervised learning, with performance improving to 98% when using semi-supervised techniques. Building upon these results, we explored a semi-supervised teacher-student training approach combining Mobilenet v2 and Xception architectures. This combined model achieved an accuracy of 70% on the test set.
Keywords
Convolutional Neural Network Weeds Classification Deep Learning Mobilenet_V2 Xception Teacher-Student Framework.
Les articles similaires
Bahi Meriem
.
pages 39-58.
Rouag Noureddine
.
Mekhlouf Abdelhamid
.
Makhlouf Mahfoud
.
Rouabhi Amar
.
pages 23-30.
Nacima Deghiche_diab
.
Samira Karoune
.
Tesnim Deghiche
.
pages 95-102.
Meguellati Farida
.
pages 229-240.
