Research on Rice Disease Identification Model Based on Migration Learning in VGG Network

Abstract

[Purpose]Aiming at the problems of slow speed, over-fitting and unsatisfactory recognition of traditional recognition models, VGG16 convolutional neural network is applied to rice disease identification, and the knowledge learned by VGG16 model is transferred to rice disease identification by means of migration learning method to construct the recognition model. [Method] The classifier uses linear discriminant analysis by setting different learning rates to fine tune the network parameters. Using image processing technology to expand the disease image set, according to the disease characteristics, the sample is divided into 10 types of diseases such as rice blast, sheath blight, rice smut and bacterial blight, and 90% of each sample is randomly selected as the training set. 10% of the samples were used as test sets, and the samples were trained using the 10-fold cross-validation method. During the simulation training, the other two methods were compared and analyzed. [Result] The experimental results show that the accuracy, balance F score and prediction speed of the disease recognition model using parameter trimming and linear discriminant classifier are higher than the other two recognition models. The average correct rate is 97.19%, the equilibrium F score is 96.75%, and the prediction speed is 175 samples per second; from the performance analysis of the model, the accuracy rate did not show large fluctuations. After 100 trainings, the training and prediction accuracy rate reached 96%, the loss rate showed a gradient downward trend, and the change was relatively stable; The 10 types of disease samples were predicted, and the average accuracy was 97.08%, the recovery rate was 96.22%, and the F1-score value was 96.75%. [Conclusion] The method has the characteristics of high accuracy, strong generalization ability, good robustness and small loss rate. This provides reference and reference for plant disease identification research.