Rice Pest Prediction Based On Bi-LSTM Model

In the development of modern agriculture,crop diseases and insect pests are important factors restricting agricultural output and improving agricultural development,the prevention and prediction of diseases and insect pests has become the focus of more and more experts and agricultural workers.With the rapid rise of deep learning and big data,there are more and more methods for prediction of diseases and insect pests at home and abroad,but improving the accuracy of prediction of diseases and insect pests by various means and strengthening the prevention and control of diseases and insect pests are still the main research directions of the current crop industry development.Research has found that there is a certain relationship between crop diseases and insect pests over the years and changes in local historical weather conditions,and some rules can be drawn from these data changes over time.Many scholars use long and short-term memory networks(LSTM)that have certain advantages in dealing with time series problems,has carried out the prediction of pests and diseases,and has achieved good research results,the bidirectional long short-term memory network(Bi-LSTM),which is a variant of the LSTM model,is rarely used in such problems.This model is widely used in the field of natural language processing and also has certain advantages in dealing with time series problems.Based on the above considerations,this paper attempts to use the Bi-LSTM model to predict the disease and insect pests of rice.The main researches are as follows:1.Statistical and correlation analysis on the acquired original data set.Firstly,more than 18,000 pieces of data downloaded from the pest and disease system are sorted and classified,including rice pest and disease data and eight weather data in different regions over the years.Correlation analysis was carried out,and it was found that the highest temperature,lowest temperature,relative humidity in the morning,relative humidity at night and pests were highly correlated,and these data were related to time changes.It was decided to use these five types of factors as the predictors of this model.2.Use the selected data set to train the Bi-LSTM model to obtain the optimal model parameters suitable for this topic.Firstly,through the analysis of the data and the problem of disease and insect pest prediction,the activation function suitable for the model and the four network parameters that have a great influence on the accuracy of the model are determined.Then four training sets are selected as input,and the method of controlling a single variable is used to control the model’s performance.Each parameter is adjusted and trained,and the parameter value with the best training effect is selected to build a model,so that the prediction effect of the adjusted model is optimal.3.Use the adjusted model to predict rice diseases and insect pests.This experiment made predictions on three aspects of rice diseases and insect pests,namely,the occurrence of rice diseases and insect pests(no disease and insect damage,occurrence of diseases and insect pests),prediction of rice diseases and insect pests at different time periods in the future(one week,two weeks,three weeks,four weeks in the future),Prediction of the occurrence degree of rice pests and diseases(severe,normal,minor).The data sets used in the first two predictions are the same.For the prediction of the occurrence of rice diseases and insect pests,this paper re-uses different data sets to adjust the model parameters,and finally selects the parameter values with the best prediction results to build a model for prediction.The final experimental results show that the Bi-LSTM model has good performance in three aspects of rice disease and insect pest prediction.4.The prediction results of using the recurrent neural network RNN and LSTM models are compared with the prediction results of Bi-LSTM.At the end of this paper,the prediction results of the model are compared with the prediction results of the other two models.The comparison results show that the prediction effect of the optimized Bi-LSTM model is better than the other two models,and has achieved better prediction performance,indicating that the Bi-LSTM model has better prediction performance.The LSTM model also performs well in the prediction of crop diseases and insect pests.To sum up,the model can provide reference value for practical research in the field of crop disease and insect damage prediction.