Table 5 Comparison of classification accuracy among our models and two benchmark CNN models

From: Open-set convolutional neural network for infrared spectral classification of environmentally sourced microplastics

Model

Accuracy

Notes

Our SoftMax model

97.3%

Evaluated on the full Test Set I

Our OpenMax model

96.0%

Evaluated on the full Test Set I

Closed-set 1D CNN (pretrained)

47.6%

Evaluated only on Test Set I samples belonging to classes known to the model

Closed-set 2D CNN (reproduced)

56.4%

Evaluated only on Test Set I samples belonging to classes known to the model

  1. Test Set I refers to the subset of the test set containing spectra from the 18 known classes included in this study.
  2. Closed-set 1D CNN refers to the model of Liu et al.23; we used their released pretrained weights. Closed-set 2D CNN follows Zhu et al.35; since pretrained weights were not provided, we reproduced the model from the authors’ released code and data. Our implementation was faithful to the originals, using the same software environment (e.g., Python/MATLAB), preprocessing, training/testing settings.
Back to article page