基于轻量级残差网络的苹果叶病识别

Lightweight Residual Networks for Diagnosis of Apple Leaf Diseases

【目的】解决卷积神经网络在复杂环境下识别率低、模型参数多等问题,为苹果叶病智能识别提供参考。【方法】本研究提出一种基于改进ResNet18的苹果叶病识别模型。首先,通过离线增强和在线增强两种方式解决数据不平衡和数据过拟合现象,增强模型的泛化能力;其次,引入缩放因子调整通道参数以减少网络参数量,并在下采样残差结构的恒等映射中用最大池化层代替1×1卷积完成下采样,去除图片中的冗余特征,增大模型的感受野;将ResNet18模型的第一层7×7卷积层替换为多尺度特征提取模块,提高模型对细小病斑的提取能力;最后,在特征提取网络中插入DenseBlock模块,加强模型对浅层有效特征的重用。【结果】改进后的ResNet18模型准确率为97.94%,比原模型高出0.88个百分点;模型大小为3.97MB,比原模型减小90.77%。与ShuffleNetv2、MobileNetv3、EfficientNet等轻量化模型和Inceptionv2、DenseNet、ResNet等经典模型相比,该模型拥有更好的性能。【结论】改进后的模型在复杂环境下能够准确识别苹果叶病,并且具有较少的模型参数,方便移植到移动设备上,为苹果叶病的智能诊断提供参考。

【Objective】An improved diagnostic model on apple leaf diseases was developed applying the lightweight residual networks.【Method】Data imbalance and overfitting of the original ResNet18 model was reduced by offline and online enhancements on its generalization ability.A scaling factor was introduced to minimize the number of network parameters and maximize the pooling layer in the constant mapping of down-sampled residual structure instead of 1×1 convolution.Redundant features in pictures were eliminated,and sensory field of the model increased.The first 7×7 convolutional layer of ResNet18 was replaced with a multi-scale feature extraction module to enrich fine lesions extraction.Finally,the DenseBlock module was inserted in the network to fully utilize valid shallow features.【Results】The improved ResNet18 model achieved an accuracy of 97.94%,which was a 0.88 percentage increase,with a significant 90.77%reduction on the program size of 3.97 MB.It performed superbly in comparison to other light-weight models,such as ShuffleNetv2,MobileNetv3,and EfficientNet,or the classical models,such as Inceptionv2,DenseNet,and ResNet.【Conclusion】The improved ResNet18 model could accurately identify the apple leaf diseases under complex circumstances.With fewer parameters than the original program,it could be more easily installed in a variety of devices for the diagnosis.