Fig. 6 | Scientific Reports

Fig. 6

From: Deep learning classification of INSV-associated weeds in Monterey county using a curated RGB image dataset

Fig. 6

F1 scores for the augmented 10 image classes across three neural networks: (a) DenseNet-121, (b) ResNet-101, and (c) ResNet-50. All models show strong class-level performance, particularly for high-fertility and drought conditions. ResNet-101 achieves the most consistent F1 scores across classes, while DenseNet-121 exhibits slightly more variability, especially for the overwatered (OS) and injured (IS) sowthistle classes. These results highlight the class-specific predictive strengths of each architecture when trained with data augmentation.

Back to article page