This study examines Convolutional Neural Networks (CNNs), which are specialized deep neural networks designed for processing data. It aims to evaluate how well various CNN architectures, including MobileNetV2, InceptionV3, and Xception, perform in image classification tasks. The dataset used in this paper contains 1,081 images categorized into 6 classes. To choose the best architecture for our research, a comparative study was conducted to evaluate different architectures of CNNs. Through detailed conducting tests and evaluation, this research aims to shed light on the effectiveness of different CNN architectures. InceptionV3 achieved a validation accuracy of 92.78%, accurately classifying around 92.78% of the validation images. MobileNetV2 outperformed slightly with a validation accuracy of 93.51%, correctly classifying approximately 93.51% of the validation images. The initial training of Xception yielded a validation accuracy of 92.58%, like InceptionV3. But, after further optimization with Adam, the final Xception model significantly improved its performance, achieving a validation accuracy of 94.37%.