基于改进ConvNeXt的大豆叶片病害分类研究

Soybean Leaf Diseases Classification Method Based on Improved ConvNext

针对现有的卷积神经网络在大豆叶片病害分类中存在复杂背景干扰的问题,提出一种改进的ConvNeXt算法,并对大豆两种常见病害以及健康叶片进行分类识别。通过在传统ConvNeXt算法的基础上增加多个注意力模块,使网络更能关注具有辨别性的特征,并选用LeakyReLu激活函数代替ReLu激活函数,避免神经元失活的现象。此外通过对数据集进行数据增强,操作多样化病害数据集,提升网络的鲁棒性。结果表明:改进的ConvNeXt算法对大豆叶片病害在测试集上的平均分类准确率均优于原ConvNeXt、ResNet50以及Swin Transformer 3个对比模型。在数据增强后,测试集上的平均识别准确率达到85.42%,研究结果可为解决复杂背景信息干扰情况下大豆叶片病害图像分类提供参考。

To solve the complex background interference problem of existing convolutional neural networks in soybean leaf disease classification,an improved ConvNeXt algorithm is proposed,and two common diseases of soybean as well as healthy leaves are classified and identified.By adding multiple attention modules to the traditional ConvNeXt algorithm,the network is made more capable of focusing on discriminative features and the LeakyReLu activation function is chosen instead of the ReLu activation function to avoid the neuron deactivation phenomenon.In addition,the robustness of the network was improved by performing data enhancement on the dataset to diversify the disease dataset.The results showed that the improved ConvNeXt algorithm outperforms the original ConvNeXt,ResNet50 and Swin Transformer,all three comparison models,in terms of average classification accuracy on the test set.The average recognition accuracy on the data enhanced test set reached 85.42%,and the research results can provide reference for solving the classification of soybean leaf disease images under complex background information interference.