Research And Application Of Wheat Cold Resistance Recognition Based On Deep Learning

Wheat is the main grain crop and important strategic Grain Reserve in China.The stability of its output is of great significance to ensure national food security and meet market demand.The irresistibility of extreme cold wave weather puts forward higher requirements for cold resistance of wheat.At present,the identification of wheat cold resistance mainly depends on the accumulation of previous experience,the measured data in the experimental field and the analysis of physiological and biochemical aspects,and the overall efficiency is low.With the development of big data technology,it is possible to obtain a variety of characteristic information of wheat;the development of information technology and artificial intelligence technology helps to obtain more effective information from the massive wheat characteristic data to identify the cold resistance of wheat.In this paper,the identification research of wheat cold resistance based on deep learning was mainly carried out,and the identification research and application of wheat cold resistance were completed from three aspects: constructing wheat germplasm corpus,training wheat cold resistance identification model based on deep learning,and designing wheat cold resistance identification application system.The main work of this paper includes the following three aspects.(1)The database of wheat germplasm resources was constructed.Modern breeding technology needs to build crop germplasm resources.At present,the research on wheat germplasm resource bank mainly focused on wheat molecular,wheat gene,and their regional traits and quality traits,lacking a unified database system for national wheat characteristic data.In this paper,we used the information of national and provincial wheat characteristics,combined with the data accumulated by wheat breeding experiment center,to build wheat germplasm resources database.The software of wheat germplasm resource database was designed and developed.After expansion,the number of wheat varieties and characters in the database reached 3049 and 38 respectively.Through data analysis,the correlation between different characters of wheat was obtained,and the distribution of plant height and seedling characteristics in wheat cold resistance grade was analyzed.(2)The research of Wheat cold resistance recognition methods based on deep learning.From the perspective of text classification of natural language processing,we use the text data from the text database of wheat germplasm resources to train the text classification model of wheat cold resistance,and use the results of text classification to realize the recognition of wheat cold resistance.Firstly,convolution neural network combined with recurrent neural network(CNN-RNN)was used to realize four classification of wheat cold resistance recognition.Compared with the single CNN and RNN methods,the accuracy,F1 value and kappa coefficient of CNN-RNN are 78.65%、53.99%和 0.5868,respectively.Secondly,multi feature dynamic fusion method was used to realize two classification of wheat cold resistance recognition.After fusion of multiple features,the accuracy and kappa coefficient of winter wheat cold resistance recognition reached 91.13% and 0.7911 respectively.Finally,XGBoost and Ada Boost combined feature selection method was used to select wheat character combination.Based on the four classification models of wheat cold resistance identification,the performance of different character combinations on wheat cold resistance identification was verified,and 15 characteristics which had a great impact on wheat cold resistance identification were obtained.(3)The recognition system of wheat cold resistance was designed and implemented.Wheat cold resistance identification is a time-consuming and complex work.According to the functional requirements of wheat cold resistance identification system,the system is composed of four parts: wheat germplasm resource database,word vector generation,model training and cold resistance prediction.Qt development tool is used to design system interface and realize human-computer interaction.The system design can effectively reduce the time consumption and complexity of wheat cold resistance identification.