Prediction Of Cotton Pests And Diseases Based On Long Short Term Memory Network

With the development of computer science and the advent of cloud computing and big data era,machine learning and deep learning methods have been widely used into various fields,and they have achieved initial success in the agricultural field.The Long Short Term Memory(LSTM)network is a neural network that is based on the traditional Recurrent Neural Network(RNN).The improved LSTM not only inherits the advantages of RNN in dealing with time series problems,but also solves the gradient disappearance and gradient explosion problems that RNN often appears.The model has long-term memory function.This paper used LSTM to predict the occurrence of cotton pests and diseases.Through continuous training and iterative optimization of LSTM network,it finally achieved good prediction results on the prediction of cotton pests and diseases.This paper collected cotton pest and disease data,weather factor change data and partial atmospheric circulation data in India from 1981 to 2011,and constructed a prediction model of pest occurrence using LSTM network.The main contents of this article are summarized as follows:1.Statistical analysis of the intrinsic link between weather and cotton pests and diseases.This paper counted the annual average trends of eight weather factors and cotton pest occurrence data from 1981 to 2011.The results showed that the degree of pests and diseases has increased year by year,and the degree of damage has increased in 2005-2011.At the same time,the temperature,humidity,and rainfall in the Indian region were decreasing year by year,contrary to the trend of pests and diseases.Then,this paper analysised the correlation between the year,month and weather factors in the data and the occurrence of pests and diseases.The results showed that the occurrence of cotton pests and diseases in different regions and different types of cotton was significantly correlated with weather factors.Among them,the three factors of temperature,humidity and daytime evaporation were more common.The general law of the occurrence of weather and pests and diseases provided a theoretical basis for the prediction of the occurrence of pests and diseases in this paper.2.Two-class prediction of cotton pest occurrence based on LSTM.The weather-cotton pest and disease occurrence data obtained in this paper is time-series data,and there are certain rules to followed in the occurrence time.As an improved RNN,LSTM was well suited for modeling such problems.Firstly,the cotton pest occurrence data and eight weather characteristics data from 1981 to 2011 in India were downloaded from the crop pest control decision support system,and simple data cleaning and pretreatment were carried out to meet the basic requirements of’modeling input.Then,the important parameters of the LSTM network are set using the single variable principle.Finally,using historical weather-cotton pest and disease occurrence data to predict the occurrence of cotton pests and diseases in various regions of India in the future.To highlight the advantages of LSTM in dealing with long-term dependency problems,traditional machine learning models were used to compare against this model.The results showed that LSTM is superior to traditional machine learning models in various performance indicators.3.Multi-classification prediction of cotton pest and disease damage based on LSTM.Based on the original data,this paper divided the degree of damage of cotton pests and diseases in India into four levels:no pests,minor pests,moderate pests and diseases and serious pests and diseases.Af’ter reviewing a large amount of literature,it is found that part of the atmospheric circulation index can further influence the occurrence of local crop pests and diseases by affecting climate change in various regions.In addition,this paper downloaded some atmospheric circulation index from the National Climate Center of China Meteorological Administration as a supplementary feature,and used it together with weather factors for LSTM network modeling.Subsequently,the LSTM model was built and data prediction was performed.At the same time,data on cotton pests and diseases in different regions and different types were tested to test the universality of the model.Finally,the paper also tested the performance of the model for different prediction time lengths.The results show that this model not only predicted the occurrence of pests and diseases in the next week,but also provided a reference for the occurrence of cotton pests and diseases in the next two weeks or even one month.