Study On Nitrogen Detection Method And Application Of Greenhouse Tomato Based On Multi-source Light Information

China is a large country of tomato planting and consumption,and the planting area of facility tomato accounts for about 57%of the total area.However,the investment in facilities is low,and there is a lack of feedback of crop information in greenhouse water and fertilizer and environmental regulation production management in our country.With the development of agricultural science and technology,the combination of facility agriculture and modern information processing technology to efficiently obtain the nutrition,growth and other information of facility crops has become one of the effective ways for the modernization of facility agriculture.Based on the terahertz spectroscopy,hyperspectral and optical field imaging techniques and combined with various data processing algorithms and chemical measurement methods,this paper establishes a multi-level model for detection of nitrogen content in facility tomatoes,and designs a detection platform applicable to facility crops.The main contents of this paper are presented as follows:(1)Tomato samples with different nitrogen stresses were cultured using perlite as the substrate by adjusting the ratio of nutrient solution.The terahertz time-domain spectroscopy system,the hyperspectral measurement system and the optical field camera were adopted to obtain the spectral and optical field.The measured value of tomato nitrogen was determined by physicochemical analysis.(2)The collected terahertz spectral data were denoised by combining the S-G smoothing algorithm and the cubic spline interpolation method.The sample set was divided using KS,RS and SPXY algorithms,and the SPXY algorithm was priority selected.The pre-processed terahertz spectral data were filtered using i PLS,SCARS and UVE algorithms.Based on the selected characteristic frequency bands,a THz-based detection model of tomato nitrogen content was established by using RBFNN and BPNN algorithms.The results showed that the accuracy of the model established by RBFNN algorithm is slightly better,with RMSEC=0.1322%,RMSEP=0.1855%,,R_c~2=0.8714、R_p~2=0.8463.(3)WT smoothing algorithm and MSC correction algorithm were combined to pre-process the measured hyperspectral data,and SPXY algorithm was used to divide the sample set.To improve the efficiency and accuracy of the model,VISSA,SAE,and SPA algorithms were adopted to screen multiple sets of hyperspectral feature wavelengths,respectively.With the combination of the screened feature wavelengths,a hyperspectral-based tomato nitrogen content detection model was developed using RBFNN and BPNN algorithm.As for accuracy,the model established by the RBFNN algorithm had better detection results,with RMSEC=0.1642%,RMSEP=0.1746%,R_c~2=0.8517,and R_p~2=0.8431.To combine the advantages of terahertz spectroscopy and hyperspectroscopy in nutrient detection of facility crops,fusion models were developed in the data and feature layers.The results indicated that the accuracy of the data-level fusion model is lower than that of the single feature dimension model and the feature-level fusion model because of some redundant information in the data,while the feature-level fusion model based on the RBFNN algorithm has the highest accuracy,with RMSEC=0.1179%,RMSEP=0.1186%,R_c~2=0.9152、R_p~2=0.9106.(4)According to the characteristics of the light field data,k-means clustering and ASBFS refinement algorithm were adopted to preprocess the 3D point cloud data of tomato samples,respectively.The plant height parameters were obtained by fitting the point cloud data of the basal environment of tomato samples.Based on the idea of differentiation,the preprocessed light field data were processed by adopting semantic segmentation and slicing method,and the cut point cloud edge contours were derived by using search sorting,so as to obtain the volume and stem width parameters of the sample point cloud.On the basis of sample point cloud growth parameters,the fitting and multiple regression models of growth and nitrogen content were established respectively.To fit the actual needs of information detection of facility crops,an information detection platform applicable to greenhouse environment was designed and tested for control accuracy.This study provided a new theoretical method for rapid detection of nitrogen content and growth of greenhouse tomato,and laid a foundation for efficient and fine management of greenhouse tomato.