The Study Of Cotton Main Agronomic Characters By Image Recognition

This study aimed at establishing models and methods to monitor cotton main agronomic characters which are obtained in cotton fields by analyzing the correlation of the color characteristic of digital images and cotton population parameters.By doing so,several population features have been acquired,such as chlorophyll,nitrogen,coverage,leaf area index,biomass,regularity degree,height of plant and so on.The goal is quickly and accurately diagnosing cotton growing status.Acquired advances are as follows:1.The correlation between color parameters of cotton canopy digital image and chlorophyll content of cotton functional leaf and population greenness index were analyzed.The results showed that the correlative relationship of the color characteristics such as G-R,(G-R)/(G+R),r/g,g/r and g-r in the RGB color system,and Hue in the HIS color system with chlorophyll content of functional leaf reached significant correlation.There was a significant correlation between the population greenness index and color parameter.The chlorophyll predicted models were established.The tested results about the regression models induced that G-R was the best parameter to monitor cotton population chlorophyll information.The relative error of chlorophyll content estimation was about 6.96%and 11.60%.The chlorophyll content (CHL.C)predicted model was CHL.C=-1.3008+0.2125(G-R)-0.0038(G-R)²(R²=0.8669^**),and population greenness index(PGI)predicted model was PGI=-0.9726+0.1227(G-R)-0.0016(G-R)²(R²=0.7487^**).2.The study found that using the color characteristics in the RGB color system to monitor nitrogen status of cotton population is better than using Hue in the HIS color system by analyzing the correlation of the color characteristic of cotton digital images and nitrogen content.The precision of predicting nitrogen content showed that the nitrogen content of leaf,plant and non-leaf organs is descending in different organs. This paper put forward taking the population nitrogen index as the monitoring index of the nitrogen nutrition status and it improved the precision obviously.The tested results about the regression models induced that G-R was the best parameter to monitor cotton population nitrogen information.The nitrogen content of leaf predicted model was NCL=1.0344+0.0810(G-R)(R²=0.7171^**),and population nitrogen index of leaf predicted model was PNIL=-0.4336+0.1007(G-R)(R²=0.8654^**).The predicted precision was about 88.89%and 89.07%.3.Cotton population image is segmented in order to acquire vegetation coverage,which can reflect the size of cotton canopy.Acquirement of regularity degree is based on managing vegetation coverage.One cotton image is divided uniformly into 16 areas.Work out vegetation coverage of these 16 areas one by one and then choose an arithmetic applying to those data.This regularity degree is significant correlation to the regularity degree of height and width of plant.4.The study found that using the color characteristics to monitor canopy information index is better than to monitor leaf area index of cotton population and G-R was the best parameter to monitor cotton canopy information index.The predicted model was CⅡ=-0.4249+0.0882(G-R)(R²=0.5970^**),and the right percent was about 81.95%.There was not a significant correlation between the color parameter of cotton digital images and ground fresh biomass,but ground fresh biomass density is significant correlation to color characteristics.The tested results about the regression models induced that r/g was the best parameter to monitor ground fresh biomass density.The predicted model was GFBD=-182805.5210 +256694.0099(r/g)(R²=0.5143^**),and the right percent was about 78.82%.5.Measuring distance,such as row spacing and height of plant,could be extract by image recognition technology.In comparison with manual measurements,the accuracy rates of measurements by software were beyond 90%.The water of soil has a influence on the growing of cotton,and the color parameters reflect the water information in cotton field.As a result,there was a significant correlation between image color characters and water content of soil that is between two mulches.The results show that technology based on digital image processing technology can work successfully in the field of crop science.Population features have been acquired and analyzed in an efficient,convenient, real time,rapid and nondestructive way,contrasted with traditional manual methods.