| Potato is a global economic crop and plays an important role in agricultural production.Leaf chlorophyll content and its changes can well indicate the growth and health of plants.At present,the traditional chlorophyll detection methods have low efficiency and high operational requirements,and do not match the requirements of rapid and accurate crop growth information acquisition in refined agriculture.Therefore,developing a method that can detect crop chlorophyll in real time has become a current research hotspot.Hyperspectral imaging technology has the characteristics of high resolution,image spectrum unification,high number of spectral bands,and can perform long-term dynamic monitoring of target objects,it can be used as a new type of crop chlorophyll detection method.Based on hyperspectral imaging technology,the thesis carried out research on the non-destructive and rapid detection of potato leaf chlorophyll.According to the potato leaf chlorophyll prediction model,the leaf chlorophyll distribution map was obtained by using image processing technology.In this paper,the relationship between chlorophyll distribution map and potato disease detection was established,in order to achieve the purpose of hyperspectral imaging technology for simultaneous detection of potato leaf chlorophyll and disease.The hyperspectral imaging system is used to collect the hyperspectral image data of healthy potato leaves,and the image data is subjected to spectral preprocessing such as reflectance correction,mask,standard normal variable transformation.Use correlation coefficient,continuous projection algorithm,genetic partial least square method to obtain 5 groups Hyperspectral characteristic parameters: The first derivative of the spectrum and the correlation coefficient get five bands of 498,612,685,742,934 nm.The second derivative and correlation coefficient get two characteristic bands of 496 and 617 nm.Combining vegetation index with correlation analysis to get MCARI,TCARI,MTCI,CARI.The successive projections algorithm was used to obtain 8 characteristic bands,and the genetic partial least squares algorithm was used to determine 10 characteristic bands.The chlorophyll content of potato leaves was determined by spectrophotometry.Use general function method,improved neural network,partial least square regression,and least square support vector machine regression to model potato chlorophyll content.According to the coefficient of determination and root mean square error,the five optimal models under the above modeling method are obtained.This article focuses on the comparative analysis of the modeling effect before and after the improvement of the neural network: Under the same conditions,the improved neural network model has a determination coefficient of 0.8978 and a root mean square error of 0.0834mg/g.The coefficient of determination of the ordinary neural network model is 0.9472,and the root mean square error is 0.216 mg/g.The results show that the neural network improved by Bayesian regularization in this paper has higher prediction accuracy and lower over-fitting phenomenon,and it is used in the rapid nondestructive detection of potato leaf chlorophyll with other optimal models.Using the optimal chlorophyll prediction model in this paper,the distribution of chlorophyll on potato leaves is obtained through calculation and image processing.Experimental results show that the lowest value of chlorophyll of healthy potato leaves is above 0.9mg/g,the highest value of chlorophyll near the main leaf vein,and the lower chlorophyll of the leaf edge.The paper finally explored the feasibility of using the quantitative data,color changes and "holes" in the chlorophyll distribution map to determine whether the potato is diseased,and provide theoretical support for the subsequent use of the chlorophyll distribution map to diagnose potato diseases. |
