基于神经网络算法的水果分类技术及其实现

Fruit Classification Technology and Its Implementation Based on Neural Network Algorithms

摘要: 本文提出一种基于CBAM注意力机制改进的轻量级MobileNetV2水果分类系统。首先，以MobileNetV2为骨干网络，在其特征提取末端嵌入CBAM模块，通过通道注意力和空间注意力增强模型对水果关键判别区域的特征表达能力；其次，在训练过程中动态引入MixUp和CutMix混合增强策略，有效提升模型的泛化能力与鲁棒性；最后，基于Tkinter框架开发了可视化GUI演示程序。实验在10类常见水果数据集上进行，结果表明，改进后的模型在保持轻量化的同时取得了良好的分类准确率，验证了所提方法的有效性。

Abstract: This paper proposes a lightweight fruit classification system based on an improved MobileNetV2 with the CBAM attention mechanism. First, MobileNetV2 is adopted as the backbone network, and a CBAM module is embedded at the end of its feature extraction pathway to enhance the model’s ability to represent key discriminative regions of fruits through channel attention and spatial attention. Second, MixUp and CutMix data augmentation strategies are dynamically introduced during training to effectively improve the model’s generalization ability and robustness. Finally, a visual GUI demonstration program is developed using the Tkinter framework. Experiments are conducted on a dataset containing ten common fruit categories. The results show that the improved model achieves good classification accuracy while maintaining a lightweight architecture, demonstrating the effectiveness of the proposed method.