| Rice is the main food crop in most areas of China.High-yielding,efficient and highquality rice cultivation and production is a guarantee for people’s sustenance,and also greatly promotes the economic development of China.In recent years,rice diseases have been frequent,seriously affecting rice production and adversely affecting the livelihood of the nation.Rice diseases are abundant and harmful,and disease identification in the traditional mode is mainly done manually by farmers,relying on the naked eye to identify the morphological characteristics of the disease.Such discriminative means are subject to subjective factors,error-prone and slow,with low recognition efficiency.With the booming development of deep learning technology,deep learning technology is widely used in the field of crop disease identification and has achieved better recognition results.Therefore,this paper uses deep learning methods and proposes a study on rice leaf disease recognition based on deep learning,using a convolutional neural network model to focus on the classification and recognition of common rice leaf disease images,further improving the recognition efficiency by optimizing the convolutional neural network model,and completing the design of the disease recognition system based on the Flask framework.The main work accomplished in this paper is as follows.(1)Collect common rice leaf disease images,construct a rice leaf disease dataset,preprocess the dataset and then fully expand the dataset using data expansion means,and divide the dataset into a training set and a validation set in the ratio of 7:3 for the training and validation of the model.(2)Analyze the effect of different convolutional neural network models on the recognition of rice leaf diseases.Six common classical convolutional neural network models were trained using the rice leaf disease dataset,and information such as recognition accuracy,recognition time and model weight size were comprehensively compared among the models.The results show that Res Net34 has smaller model weights with higher accuracy and shorter training time,which is suitable for rice leaf disease recognition research.Therefore,Res Net34 was selected as the basic network model for optimal design.(3)Since the disease features in rice leaf disease images are not obvious,the total number of data samples is small,and the background is more complex in practical application scenarios,the recognition efficiency is not good,so the Res Net34 model is optimized and studied using the optimization method based on attention mechanism and the optimization method based on migration learning.The research results show that the improvement methods selected in this paper can both improve the model performance and increase the recognition accuracy to a certain extent.(4)To design and implement a rice leaf disease recognition system to improve the use of the model studied in this paper. |
