Crop Disease Recognition Method Based On Model Ensemble

With the continuous development of society and economy,people have higher requirements on the quality and yield of agricultural products.Using traditional methods for identifying and diagnosing crop diseases has fallen short of the requirements for efficient identification of crop diseases.Crops can show symptoms through leaves.Collecting images of crop leaves and using scientific and intelligent diagnostic methods are particularly important for disease identification.However,in the process of leaf image collection,it is easy to be affected by the natural environment.The traditional image recognition method for crop disease identification cannot effectively solve the problems of these factors.Therefore,how to quickly and accurately identify and diagnose crop diseases has gradually become the agricultural field Research hotspots.In recent years,convolutional neural networks have developed rapidly in the field of image classification and recognition.Experts and scholars have applied them to crop disease image recognition and achieved good results,but there are still some shortcomings.Therefore,this paper takes crop disease images as the research object and uses multiple convolutional neural network model fusion methods to improve the existing deficiencies.The main research contents include the following three parts:1.The crop disease images collected in the field environment have the problem of uneven distribution and small amount of data of individual species.Therefore,this paper first conducts data augmentation operations on the crop data set,and then uses Focal loss function according to the characteristics of the uneven data.2.The characteristics of crop disease images are effective expressions of disease information.In this paper,four convolutional neural network models,ResNet101、ResNext50、SE-ResNet50 and SE-ResNext50,are used to extract crop leaf disease characteristics.The weights are transferred to the training of the crop disease data set,and the pre-trained model is fine-tuned to improve the classification accuracy of the data of this paper by a single model.3.In order to improve the recognition accuracy and make up for the shortcomings of a single model,this paper proposes a Stacking algorithm combining a convolutional neural network and a machine learning algorithm.The four convolutional neural network models are used as the basic model of the first layer of the Stacking algorithm,and the meta-learner XGBoost is used as the second layer of the Stacking algorithm to form the model fusion algorithm in this paper.This paper uses a model fusion-based method to identify crop disease images.After fusion,the accuracy rate is 87.19%,which is 2.59% higher than the single model of ResNet101.For the problem of data imbalance,the Focal loss function is used to reduce the error of difficult classification samples.Compared with the traditional cross entropy function,the accuracy rate is improved by 0.06%.Compared with other methods,this method has higher recognition accuracy,which shows the effectiveness of this method in crop disease identification.