Extraction And Analysis Of Phenotypic Traits Of Potted Wheat Based On Deep Learning

As an important trade commodity and food crop,wheat plays an increasingly important role in food security,and China is the country with the largest wheat planting area and the highest yield in the world.It is very important to maintain the high yield of wheat to ensure China’s food security.Plant height,stem and leaf are important factors affecting wheat yield.The morphological,color and texture characteristics of mature wheat plants were extracted,which can be used to model and predict wheat yield.However,using phenotypic traits of mature wheat can only estimate yield,but can not promote yield.Because the mature stage of wheat has been fully developed,it is impossible to control its growth and development through precision irrigation,precision fertilization and other means,so as to achieve the effect of yield increase.The growth status of wheat at seedling stage will have a certain impact on its final yield,so it is very important to accurately extract the phenotypic traits of wheat at seedling stage.By mining the relationship between the early growth and development traits such as leaf growth rate,stem growth rate and yield of wheat at seedling stage,it can provide the basis for precise water and fertilizer management of wheat at seedling stage and promote wheat yield.In this paper,the extraction and analysis of phenotypic characters of wheat are divided into two parts.One part is the extraction and analysis of potted seedling wheat,and the other part is the extraction and analysis of mature wheat.For the extraction and analysis of wheat phenotypic characters of potted seedlings,a method for extracting wheat phenotypic characters of potted seedlings based on deep learning is proposed,which is mainly based on the continuous collection of wheat seedling pictures by high-throughput phenotypic acquisition system.The deep learning model used to extract traits needs to be processed in three stages: training,testing and optimization results,and two evaluation parameters to measure the segmentation results are observed: pixel accuracy PA and intersection union ratio IOU,so as to ensure the accuracy of the model segmentation effect.The final model IOU and PA in the data set are 0.916 and 0.869 respectively,which proves that the segmentation effect of the model meets the standard.After the image is segmented accurately by this method,the phenotypic traits such as plant height,leaf length,stem length and leaf angle in wheat seedling stage are further extracted by traditional image processing methods.The linear fitting R2 of plant height,leaf length and artificial value extracted by this method were 0.9828 and 0.8735,respectively,indicating that the extraction accuracy of phenotypic traits of wheat was good.Based on these characters,derived parameters such as leaf growth rate and stem growth rate of wheat can be calculated.For the character extraction and analysis of mature wheat,because the leaves,stems and ears of wheat mature pictures are seriously blocked from each other,high-precision labeling samples cannot be obtained,and the segmentation effect of some organs of Wheat by deep learning is not good.Therefore,this paper uses the improved threshold segmentation method to segment the whole wheat plant and extract 17 phenotypic characters,It mainly includes color traits,biomass traits,morphological traits and texture traits.TPA,the phenotypic trait with the highest correlation with yield,was screened by stepwise linear regression＿W,TPA＿W and wheat yield per plant was 0.609.TPA＿W as the independent variable and yield as the dependent variable for multi model prediction analysis,and through ten fold cross validation,we can get the linear fitting results,the correlation reaches 0.617,and the error fluctuation is between 5% and 7%,indicating the accuracy of linear fitting.At the same time,the derived parameters extracted from seedling wheat were compared with TPA＿W was modeled and analyzed,and it was found that the leaf growth rate was related to TPA＿W correlation is the best,and the correlation is 0.3405,which indirectly shows that there is a certain relationship between wheat leaf growth rate and yield.Some characters of wheat at seedling stage can be used as a reference for guiding the fine management of early growth and development of wheat.