改进MobileNetV3-Small模型在番茄叶片病害识别中的应用

Application of improved MobileNetV3-Small model in tomato leaf disease identification

面对农作物病虫害识别中的挑战,特别是移动端应用对模型准确性和效率的双重要求,文章提出了一种基于MobileNetV3-Small的改进模型。在MobileNetV3-Small模型的基础上,引入了ECA(Efficient Channel Attention)注意力机制,取代了原有的SE(Squeeze-and-Excitation)模块,从而减少了模型参数量和计算成本,同时提升了对细粒度特征的捕捉能力和抗干扰性。通过在番茄叶片病害数据集上的训练,结果表明改进后的模型准确率达到了98.93%,比原模型提高了0.54个百分点,权重文件大小从17.6 MB减少到12.3 MB,减少了30%。在各项性能评估指标上,该模型均优于传统的轻量化网络和复杂模型。研究结果为移动端农作物病虫害智能识别提供了一种新的高效方案。

Faced with the challenges in identifying crop pests and diseases,especially the dual requirements for model accuracy and efficiency in mobile applications,this article proposes an improved model based on MobileNetV3 Small.On the basis of the MobileNetV3 Small model,ECA(Efficient Channel Attention)attention mechanism is introduced to replace the original SE(Squeeze-and-Excitation)module,thereby reducing the number of model parameters and computational costs,while improving the ability to capture finegrained features and anti-interference.Through training on the tomato leaf disease dataset,the results showed that the improved model achieved an accuracy of 98.93%,an increase of 0.54 percentage points compared to the original model.The weight file size was reduced from 17.6 MB to 12.3 MB,a decrease of 30%.In terms of various performance evaluation indicators,this model outperforms traditional lightweight networks and complex models.The research results provide a new and efficient solution for intelligent recognition of crop diseases and pests on mobile devices.