一种基于改进ResNet18神经网络的玉米叶片病害识别方法OA北大核心CSTPCD

In order to develop fast and efficient method for maize leaf disease identification,we proposes a corn disease identification method based on improved ResNet18 neural network using five common maize leaf diseases such as grey spot,southern rust,small spot,rust,and leaf spot as the research object.By introducing Pyramidal Convolution on the basis of ResNet18 network,multi-scale feature information can be used to improve the identification and localization ability of single leaf in the complex growth environment,to effectively accelerate the convergence of the model and significantly improve the disease identification accuracy of the model.The activation function of the residual structure was replaced with the activation function of PReLU(Parametric Rectified Linear Unit)to avoid neuronal death during model training.Experiments on the collected real maize leaf disease dataset show that the accuracy,precision,recall and F1-scores of maize leaf disease identification are improved by 1.86%,1.78%,1.78%and 1.87%compared with the original ResNet18 residual network;The parameter size of the model was reduced by 1.85%.This model can be used as an effective method to detect maize leaf diseases in complex growth environments.