📄 中文摘要
数据增强在训练卷积神经网络时被广泛应用，尤其是在图像分类任务中。通过对训练图像进行旋转、翻转或平移等变换，可以引入更多的变化，从而帮助模型更好地泛化。然而，是否更多的增强总是能提高性能是一个常被忽视的问题。研究通过不同程度的数据增强，分析其对在 CIFAR-10 数据集上训练的 CNN 模型的影响，提供了实验结果和可视化数据，以揭示数据增强对模型泛化能力的实际效果。
🧠 理解 CNN 泛化与数据增强（TensorFlow – CIFAR-10）
📄 English Summary
🧠 Understanding CNN Generalisation with Data Augmentation (TensorFlow – CIFAR-10)
Data augmentation is widely used in training convolutional neural networks, particularly for image classification tasks. By transforming training images through rotations, flips, or shifts, more variation can be introduced, helping the model to generalize better. However, the question of whether more augmentation always leads to improved performance is often overlooked. This study investigates how varying levels of data augmentation affect a CNN trained on the CIFAR-10 dataset, presenting experimental results and visualizations to reveal the actual effects of data augmentation on model generalization.