Fig. 5 | Scientific Reports

Fig. 5

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

Fig. 5

Mean cross entropy across 10 independent data splits for each neural network trained on augmented and unaugmented weed image datasets. Models trained on augmented data consistently exhibited lower entropy, reflecting more confident and well-calibrated predictions. DenseNet-121 showed the highest entropy in the unaugmented regime, suggesting overfitting and unreliable output probabilities.

Back to article page