Research On Robust Seedling Identification Method Based On Multi-feature Information Fusion

Seedling transplanting is an important form of crop production such as vegetables.Selecting robust seedlings of suitable age as the object of transplanting can greatly improve the yield and quality of crops.At present,the evaluation criteria for the quality of seedlings at the appropriate age for transplanting are relatively single,and the detection methods are mostly human experience judgments,resulting in low efficiency of the screening process and poor confidence in the results.In view of the above problems,this paper takes the tomato plug seedlings of the right age in the spring transplanting period as the research object,and through comprehensive morphological and pigment indicators,a comprehensive analysis of the growth status of the seedlings is carried out,and a more reliable and applicable quantitative evaluation standard is formulated.Based on this standard,a quantitative analysis model of spectral data and evaluation values was established by using the advantages of hyperspectral energy for non-destructive testing,which provided a theoretical basis for the rapid detection of transplanted seedling quality in the future.The main contents and conclusions are as follows:(1)Five seedling indexes of plant height,stem diameter,whole plant fresh weight,whole plant dry weight and chlorophyll were measured for 276 tomato plug seedlings during the spring transplanting period.The weight of each index was determined by the weight coefficient method,and finally two indexes with more comprehensive information and greater influence were selected according to the weight results:chlorophyll and dry weight.The results of the correlation analysis showed that the simplified seedling evaluation value composed of the two indicators could approximately represent the comprehensive evaluation value,and the correlation coefficient R was 0.92,which greatly reduced the number of indicators required for quality inspection,and could well characterize the spring seedling transplanting.The robustness of tomato seedlings during planting.(2)The hyperspectral data of each plug seedling was extracted,and the region of interest(ROI)was selected as the research object.After denoising,S-G smoothing and multivariate scattering correction(MSC)preprocessing,the effects caused by light scattering were eliminated.Spectral interference information is more available than raw spectral information.Then the Spectral-Physicochemical Value Co-occurrence Distance(SPXY)algorithm was used to divide the sample set,and the two variables of band value and evaluation value were used to calculate the inter-sample distance to maximize the representation of the sample distribution and improve the sample difference and representativeness.(3)Adopting the competitive adaptive re-weighting algorithm(CARS)and the uninformative variable elimination continuous projection algorithm(UVE-SPA)to optimize the spectral characteristic wavenumber to 58 and 104.The dimension of spectral data and the influence of redundant information on the accuracy of model establishment and analysis speed is reduced,and simplified spectral information that can better reflect spectral characteristics is obtained.(4)Finally,two linear regression methods,multiple linear regression(MLR)and partial least squares(PLS),as well as partial least squares support vector regression(LSSVR)and convolutional neural network based on U-Net model transformation(CNN),using the preprocessed spectral data and the spectral data after extracting characteristic wavelengths as the input of the model respectively,established a quantitative analysis model of the spectral data and the simplified evaluation value,and compared and optimized.The results show that the regression effect of the model based on the two characteristic bands is better than that of the model based on the whole band,and the spectral information of the UVE-SPA preprocessing method is more abundant and effective;the modeling effect of the two nonlinear models is overall better.For the linear model,the modeling effect of the CNN model is better than that of the LS-SVR model;the UVE-SPA-CNN model has the best regression analysis effect on spectral data and seedling evaluation values,and the correlation coefficient R between the modeling set and the prediction set is R They are 0.988 and 0.946,respectively,and the root mean square error(RMSE)is 0.025 and 0.075,respectively,which provides a theoretical basis for directly using spectral data to obtain the evaluation value of tomato seedlings that integrates various factors,thereby judging the robustness of seedlings.