Evaluating Deep Learning Models for Effective Weed Classification in Agricultural Images

Effective weed management is crucial for maximizing agricultural productivity and minimizing crop losses. Traditional methods for weed detection and classification often suffer from errors and delays, particularly in the absence of specialists. Deep learning technologies offer a promising alternative for automating these tasks, potentially enhancing both accuracy and efficiency. This study compares three advanced deep learning architectures, ResNet-50, EfficientNet V2, and Vision Transformers (ViT), for classifying weed species using the DeepWeeds dataset. We explore the effects of data augmentation on model performance, evaluating each model based on accuracy, precision, recall, and F1 score. Our results demonstrate that data augmentation significantly improves model performance. EfficientNet V2 achieved the highest performance across all metrics, with a peak accuracy of 0.9703. This research provides valuable insights into selecting effective architectures and training strategies for more accurate weed detection in agricultural applications.