基于DTS-ResNet的苹果叶片病害识别方法

Recognition method of apple leaf disease based on DTS-ResNet

针对苹果叶片病害识别中传统卷积神经网络识别精度较低、收敛速度较慢等问题,提出了一种基于DTS-ResNet(ResNet based on dual transfer learning and squeeze-and-excitation block)的苹果叶片病害识别方法。该方法以ResNet为基础模型,将注意力机制与残差模块相结合作为骨干网络以强化网络对重要特征信息的提取能力、提高识别准确率,并采用双迁移学习的训练方式加快模型的收敛速度。实验结果表明,所提出的方法的识别准确率达到98.73%,能够较好地识别苹果叶片病害。相较于一些传统的卷积神经网络,该模型收敛速度更快,拟合效果更好,且具有更高的识别精度。

A DTS-ResNet-based method is proposed to address the problems of low recognition accuracy and slow convergence speed of traditional convolutional neural networks in apple leaf disease recognition. The method uses ResNet as the base model, and combines the attention mechanism and residual module as the backbone network to enhance the extraction ability of the network to the important feature information and improve the recognition accuracy. In addition, it adopts the training method of dual transfer learning to accelerate the convergence of the model. The experimental results show that the proposed method has a recognition accuracy of 98.73% and can better identify apple leaf diseases. Compared with some traditional convolutional neural networks, the model converges faster, fits better and has higher recognition accuracy.