融合空间注意力机制和DenseNet的玉米病害分类方法

Maize Disease Classification Method Based on Spatial Attention Mechanism and DenseNet

针对现有机器学习方法对玉米叶片病害进行分类时存在准确率不高、分类模型参数量偏大的问题,提出了一种融合空间注意力机制和DenseNet的玉米叶片病害分类方法SA-DenseNet。将空间注意力机制加入至DenseNet网络的每一个Dense Block模块中,使改进后网络更加关注玉米叶片病害的空间特征;把网络中的ReLU激活函数替换为PReLU激活函数,以避免在网络训练过程中输入为负数时导致的梯度消失问题。仿真结果表明,改进模型对玉米叶片病害分类的训练集准确率为98.77%,测试集准确率为98.96%,均优于AlexNet,ResNet50,ResNeXt三个对比网络。模型大小为67.7 MB,优于ResNet50和ResNeXt网络。

To deal with the problems of low accuracy and large quantity of parameters of the classification model when existing machine learning methods are used to classify corn leaf diseases,an improved classification method SA-DenseNet,which combines spatial attention mechanism and DenseNet,is proposed.The spatial attention mechanism is added to the DenseNet network,so that the improved network pays more attention to the spatial characteristics of corn leaf diseases.The ReLU activation function in the network is replaced with the PReLU activation function to avoid the problem of gradient disappearance when the input is negative during network training.The simulation results show that the improved method has an accuracy of 98.77%on the training set and 98.96%on the test set,which are better than the three comparison networks of AlexNet,ResNet50 and ResNeXt.The model size is 67.7 MB,which is better than ResNet50 and ResNeXt network.