基于注意力机制的轻量级卷积神经网络图像分类研究

Research on Lightweight Convolutional Neural Network Image Classification Based on Attention Mechanism

摘要: 随着深度学习在计算机领域的快速发展，曲面神经网络在图像分类任务中取得了显着的成果。然而，传统深度网络参数量庞大，计算复杂度高，难以在资源设定的移动设备上部署。本文提出了一种基于焦点机制的轻量化深度神经网络架构，旨在保持分类准确率的同时大幅减少模型参数和计算量。该方法通过改进引入实验结果表明，在CIFAR-10和ImageNet数据集上，所提出的模型相比经典轻量化网络MobileNetV2，在参数量减少35%的情况下，分类准确率分别提升了2.3%和1.8%，推理速度提升了28%，验证了方法的有效性。

Abstract: With the rapid development of deep learning in computer vision, convolutional neural networks have achieved remarkable results in image classification tasks. However, traditional deep networks have massive parameters and high computational complexity, making them difficult to deploy on resource-constrained mobile devices. This paper proposes a lightweight convolutional neural network architecture based on attention mechanisms, aiming to maintain classification accuracy while significantly reducing model parameters and computational load. The method effectively improves feature representation capability by introducing improved channel attention modules and spatial attention modules. Experimental results show that on CIFAR-10 and ImageNet datasets, compared to the classic lightweight network MobileNetV2, the proposed model achieves 2.3% and 1.8% improvement in classification accuracy respectively while reducing parameters by 35%, and inference speed is improved by 28%, validating the effectiveness of the method.