Application Of Improved Convolutional Neural Network Algorithm In Crop Disease Detection

Crop diseases directly affect the yield of crops in agricultural production,and timely and effective control of crop diseases is of great significance.The purpose of this research is to apply artificial intelligence technology in the computer field to detect plant disease images and identify plant diseases,so as to target diseased plants with medicines and reduce drug waste and pollution.This paper collects rice data and corn data for the problem of crop disease detection,and uses the rice disease and corn disease data to do the following work:(1)Data processing.The first is data noise reduction,which removes the background other than the plant leaves in the image and avoids background interference to the experiment;then expands the amount of data,adopts symmetric transformation and exchanges the RGB channels of the image in pairs and rotates the image to enrich the data Diversity can alleviate the over-fitting phenomenon of the model.(2)Construct a dense convolutional network based on transfer learning.Constructed a deep convolutional network with dense blocks,and used large sample data sets to train and learn knowledge,transfer and use the trained model to achieve rapid convergence in small sample data sets.Considering that dense convolutional networks have good feature learning and feature expression characteristics and the advantages of automatically learning crop diseases,and then the task of rice and corn disease identification,the transfer learning method can be merged with the knowledge learned on the data set.This can solve the problem of low accuracy of traditional image recognition.(3)Dense convolutional network based on migration learning is applied to crop disease detection.Effectively apply deep learning knowledge in the computer field to crop disease detection,apply digital image processing technology and convolutional neural network to build the system model,use Softmax regression algorithm to solve multi-classification problems,image processing and disease Detection,to achieve the purpose of automatically extracting features and accurately identifying diseases.This research also uses a variety of deep learning models to conduct comparative experiments,and the evaluation of multiple indicators shows that the dense convolutional network model based on transfer learning is suitable for detecting rice diseases and corn diseases.In order to verify the validity and generalization ability of the model,the model uses 10-fold cross-validation.The model can be extended to the diagnosis of various plant diseases in agricultural production,and has practical significance for the development of agriculture.