| Objective:Nitrogen element plays an important role in various growth cycles of maize,and how to achieve precise nitrogen supplementation to improve yield has always been a hot research topic in this field.Research has shown that nitrogen supplementation and planting density can be used to regulate individual plant nitrogen level.However,these studies have the following problems:the nitrogen measurement methods used are cumbersome,and the accuracy of visual judgment is low.Moreover,these works mainly focus on statistical analysis of nitrogen density interactions,and advanced nitrogen monitoring methods urgently need to be developed.With the development of intelligent sensing and deep learning application research,the collaborative use of corn images,spectral image data,and deep learning methods for nitrogen density research is expected to provide new monitoring methods and scientific understanding in this field.Based on this,this article proposes to use corn images and spectral image data to monitor,predict,and study the nitrogen content in corn leaf tissue.Methods:The experiments were conducted in the experimental field of Beiyuan New Area Agricultural College,Shihezi University(44°32’N,86°08’E)in 2021 and 2022,with the maize variety Xinyu 57.Four fertilization treatments were set up:0 kg/hm~2,150 kg/hm~2,300 kg/hm~2,and 450 kg/hm~2,represented by N0,N1,N2,and N3,respectively;Three density levels:60000 plants/hm~2,120000plants/hm~2,and 180000 plants/hm~2,represented by D1,D2,and D3,respectively.By obtaining data on corn plant height,stem diameter,ear height,dry matter accumulation,and yield,combined with mobile phone images and hyperspectral instruments,data on planting parameters and leaf spectra were obtained.The feature bands used to measure nitrogen content in the spectra were extracted using continuous projection method.On the basis of the above data,two deep learning models were proposed,namely a YOLOV6convolutional optimization algorithm based on SIFT algorithm and LSTM model based on multi head attention mechanism,to predict nitrogen content and verify the accuracy of results in the spectra of corn.Results:(1)Different nitrogen levels/planting densities have interactive effects on agronomic indicators such as maize plant height,stem diameter,dry matter,NUE,and yield.The stem diameter showed a trend of first increasing and then decreasing with planting density.Under N2 nitrogen level,the corresponding stem diameters for D1,D2,and D3 were 7.46 cm,8.3 cm,and 6.93 cm,respectively;Under the same planting density,different nitrogen levels have a more significant impact on stem diameter.The plant height is influenced by both nitrogen content and planting density.Specifically,as the nitrogen application rate increases,the corn plant height also increases.However,when the nitrogen application rate reaches a certain level,it decreases.The planting density also has a similar effect,but the effect is not as significant as the nitrogen application rate.During the tasseling stage of corn,the ear height also showed a trend of first increasing and then decreasing with the increase of planting density and nitrogen application rate.Both nitrogen application rate and planting density have an impact on dry matter during the jointing stage.From the perspective of planting density,both show a trend of D3<D1<D2,while from the perspective of nitrogen application rate,both show a greater trend of N1 or N2.The leaf height,Spad and ear height at tasseling stage also showed a trend of increasing first and then decreasing with the increase of plant density and nitrogen application rate.As the nitrogen application rate and planting density increase,both the Ci parameter and Cond value first increase and then decrease,reaching their maximum values at N2(Ci=22.35,Cond=0.176).In thenitrogen use efficiency,both nitrogen fertilizer application and increasing planting density could effectively increase the nitrogen uptake of maize,but the increasing rate was different in different treatments.The optimized condition N2D2 of the combination showed the highest corn yield,which increased by 34.52%compared to the experimental field without optimized nitrogen/planting density.(2)In order to quickly obtain the corresponding relationship between planting density,plant height,and nitrogen content,an optimized YOLOV6 convolutional neural network recognition method was introduced.Combined with actual nitrogen measurement,it was shown that image accuracy based on plant height/planting density can reach over 90%.(3)In order to further improve the accuracy of nitrogen monitoring,a recurrent neural network model was adopted,and a PCA based continuous projection method was used for spectral feature extraction and recognition.The analyzed reflection spectrum was corresponding to the nitrogen content.By manually labeling the spectral feature information,the accuracy was verified to reach 95%in practice.Conclusion:The nitrogen element results predicted by the proposed algorithm can be fed back to the corn planting process using planting images obtained from mobile phones and hyperspectral images.The actual farm test data shows that the application method in this thesis has practical value,providing important ideas and directions for the accurate nitrogen application process. |
PDF Full Text Request