Ensemble Learning Approach for Car and Tank Image Identification with Explainable AI

Rapid and accurate identification of battlefield targets, such as cars and tanks, is crucial for mastering battlefield situations. Traditional image classification methods have limitations in handling complex scenes and data quality. This study introduces a novel ensemble model combining attention depthwise and attention depthwise separable models for robust car and tank image classification. The dataset is preprocessed to remove background noise and augmented to increase the image count and diversity. The proposed model employs Grad-CAM for interpretability, focusing on critical regions on the image. The proposed ensemble model outperforms individual models, achieving an accuracy of $\mathbf{9 8. 4 3 \%}$, precision of $\mathbf{9 7. 3 1 \%}$, recall of $\mathbf{9 7. 3 9 \%}$, and f1-score of $\mathbf{9 7. 3 3 \%}$. The proposed ensemble demonstrates superior performance in car and tank identification, providing a reliable solution for military and security applications.