|
AUTHOR(S): Ashwani Kumar Aggarwal
|
TITLE |
 PDF |
ABSTRACT The deep learning models used for image classification in various applications of computer vision such as intelligent transportation systems, precision agriculture, medical imaging, and remote sensing, etc. perform well when the dataset used for training the models is clean meaning the images in the dataset are free from noise, distortion, motion blur, occlusions, and augmented data. However, in the presence of noisy data or augmented data, the performance of many deep learning models degrades significantly, resulting in false image classification. In this paper, the performance of various deep learning models in the presence of noise and data augmentation is evaluated on benchmark datasets. Monte Carlo dropout is used for uncertainty quantification and Grad-CAM is used for the visual explainability. The performance of the models is evaluated using performance metrics accuracy and uncertainty-Robustness Index (URI). Experimental results show that the method improves the robustness of the models in the presence of noisy data. |
KEYWORDS Image classification, Monte Carlo dropout, Visual explainability, Data augmentation, Uncertainty quantification |
|
Cite this paper Ashwani Kumar Aggarwal. (2025) Explainable and Uncertainty-Aware Deep Learning for Image Classification under Noisy and Augmented Conditions. International Journal of Signal Processing, 10, 39-42 |
|
