Harnessing deep learning for multi-class weed species identification in agriculture

Authors : Ebru Ergün

Pages : 251-262

Doi:10.28948/ngumuh.1495040

View : 36 | Download : 45

Publication Date : 2025-01-15

Article Type : Research Paper

Abstract :Effective identification of weed species is critical for efficient agricultural management, enabling targeted eradication and optimized farming practices. In this study, ResNet, VggNet and DenseNet were used to evaluate the performance of deep learning models in accurately classifying different weed species. The dataset consisted of high-resolution images of different weed species taken under different environmental conditions. The experimental results demonstrated the ability of these models to identify multiple weed species with high accuracy. Evaluation metrics, accuracy, precision, recall and confusion matrices, validated the effectiveness of the models in discriminating between species. Of the convolutional neural network architectures tested, VggNet showed the highest classification accuracy of 99.21%. The results underscored the potential of deep learning-based classification systems in advancing scalable and efficient weed species identification and management for agricultural applications.
Keywords : Tarımsal yönetim, Sınıflandırma, Derin öğrenme, Çoklu sınıf, Hassas tarım, Yabancı ot türlerinin tanımlanması