Image classification is a vital research direction in computer vision all over the world. Before the advent of deep learning, image classification relied on manual feature extraction and conventional machine learning algorithms. However, Convolutional Neural Networks (CNNs) revolutionized this field by automatically learning features from data. The article discusses the fundamental principles of convolutional neural networks and compares various CNN architectures. Key layers such as convolutional, pooling, activation, fully connected, and dropout layers are explained in detail, along with techniques like backpropagation and optimization algorithms. Additionally, common CNN models like LeNet, AlexNet, VGGNet, GoogLeNet, ResNet, SENet, and EfficientNet are introduced, highlighting their characteristics and applications.