In this paper, the performance of two deep learning models, VGG19 and MobileNetV2, on a custom dataset created specifically for this classification task. Both models achieved impressive accuracy, demonstrating their effectiveness. However, a detailed analysis of the precision and recall values of the metrics reveal key differences in their classification behavior, which are thoroughly discussed. A major contribution of this work is the development of a custom dataset with dynamic background that enhances the robustness of the models. This paper also highlights the architecture and efficiency of VGG19 as well as MobileNetV2, where latter being a lightweight, performs on par with VGG19.