Fig 1.
Sample images of various classes of mosquitoes of the comprehensive dataset used in this work.
Table 1.
Number of training images of various classes of mosquitoes of the employed dataset.
Table 2.
Dataset distribution for training, validation, and test sets.
Fig 2.
Architecture and Blocks of the Swin Transformer.
Fig 3.
Architecture of CVT-13 technique employed in this work.
Fig 4.
Architecture of MobileViT model applied in this work.
Fig 5.
A detailed workflow diagram of the proposed mosquito classification system.
Table 3.
Performance metrics of the applied models for closed-set learning.
Fig 6.
Training and validation accuracies and losses vs. epochs of the applied Swin-B and MobileViT models.
Fig 7.
Training and validation accuracies and losses vs. epochs of the applied ViT Base and Xception models.
Fig 8.
Normalized confusion matrices of the applied Swin-B and MobileViT models using the test set.
Fig 9.
ROC curves of Swin-B and ViT models for closed-set learning using the test set.
Fig 10.
Classifications of the Swin-B model with actual and predicted classes for closed-set learning on the test set images.
Fig 11.
Multiple mosquito detection by the proposed Swin-B model.
Table 4.
Performance Metrics of Models on Closed-Set Without Data Augmentation.
Fig 12.
Wrong classification of unknown insects in static setting (prior to implementing open-set learning).
Fig 13.
Weibull distribution for Swin-B model containing all mosquito classes on the extended test set images.
Fig 14.
ROC curves of Swin-B and ViT models for open-set learning with OpenMax using the extended test set images.
Table 5.
Performance metrics of the applied models for various OpenMax thresholds.
Fig 15.
OpenMax threshold vs. accuracy for various models.
Fig 16.
Classifications of the Xception model for open-set learning with OpenMax.
Table 6.
Comparison of the proposed mosquito classification system with related studies.