Establishment Of System For Monitoring Cotton Growth Based On Computer Vision Technology

Computer vision technology has been well developed and is widely used tomonitor crop growth and to diagnosis crop N status. An important research goal is toestablish a system that combines computer vision technology and near ground remotesensing to monitor crop growth and N status. The system, which should be convenient,efficient, practical, and widely applicable, could provide a new theoretical basis andtechnical support for crop monitoring.In this study, color images of cotton canopies were captured with a digital camerafitted with a charged-coupled device (CCD) as an image sensor. An image analysisapproach was developed to extract the RGB features of the images. The objectives ofthis study were to calibrate models describing the relationship between the spectral andbiological properties of cotton canopies, to verify the models using data collected fromthree fields of high-yield cotton, and to build a remote service system platform formeasuring the growth and N status of cotton based on computer vision technology.The following results were obtained from this study:1. Temporal changes in color information extracted from canopy images of cottongrown under different N rates.The two cotton cultivars used in this study, Xinluzao43(XLZ43) and Xinluzao48(XLZ48), are commonly grown in Xinjiang Province. The study, which was conductedin2010and2011, included five N treatments. Images of the cotton canopies werecaptured using a digital camera. The color parameters (R, G, B, H, I, and S) of thecotton image were extracted using a digital image recognition system (DIRS). Theresults indicated that the R and G values of the RGB model, and the I value of the HISmodel can describe cotton growth and development. The fitted curve function wasy=a-b×ln(x+c). Therefore the values of R, G, and I can be used as variable parametersto monitor cotton growth. The B and H values in the RGB and HIS models can alsodescribe cotton growth and development. The fitted curve was a quadratic polynomial:y=ax2+bx+c. The fitting parameter values of B fluctuated by a large amount, dependingon N rate. Furthermore, the variation of the B value was not significant. Therefore, theB values cannot describe cotton population growth or development. The temporalchanges in saturation (S) followed no pattern.2. Model to estimate cotton growth and N status based on canopy cover.We extracted parameter values from images of cotton canopies using digital imagesegmentation. Two cotton cultivars were grown using five N rates. The images, which were taken at different times during the growing season, where divided into twocategories: canopy and soil layer. The images were further divided into four layersbased on threshold values of the color features. The four layers were sunlit canopy (SC),shaded canopy (ShC), sunlit soil (SS), and shaded soil (ShS). In order to minimizeerrors during image processing, we obtained canopy cover (CC) values throughMATLAB image processor software and VC++computer language program. The mainobjective of this study was to develop a non-destructive method for monitoring cottongrowth and N status using a digital camera. Digital images were taken of the cottoncanopies between emergence and full bloom. The green and red values were extractedfrom the digital images and then used to calculate canopy cover. The values of canopycover were closely correlated with the normalized difference vegetation index and theratio vegetation index measured using a GreenSeekerTMhandheld sensor. Models werecalibrated to describe the relationship between canopy cover and three growthproperties of the cotton crop (i.e., aboveground total N content, leaf area index, andaboveground biomass). There were close, exponential relationships between canopycover and the three growth properties. The relationship between canopy cover andaboveground total N content was the most precise (R2=0.978; RMSE=1.479g m-2).The models were validated using three high-yield cotton cultivars. The results indicatedthat the best relationship between canopy cover and aboveground total N content had anR2value of0.926and an RMSE value of1.631g m-2. In conclusion, digital camerashave good potential as a near-ground remote assessment tool for monitoring cottongrowth and N status.3. A model to estimate cotton growth and N status based on different colorparmeters.Total N content, LAI, and aboveground biomass are the dominant growthproperties of the cotton crop. The color parameters and values of canopy images varydepending on the N status of the cotton crop. We analyzed the color parameters (i.e.,G-R,2g-r-b, and G/R) of the cotton canopy and then examined the correlation betweenthese parameters and the three growth properties mentioned above. The results showedthat G-R,2g-r-b, and G/R were highly significantly correlated with total N content, LAI,and aboveground biomass. For G-R, the correlation coefficients were0.945**for total Ncontent,0.968**for LAI, and0.935**for aboveground biomass. For2g-r-b, thecorrelation coefficients were0.906**for total N content,0.935**for LAI, and0.898**foraboveground biomass. For G/R the correlation coefficients were0.859**for total Ncontent,0.889**for LAI, and0.892**for aboveground biomass. Models wereestablished to describe the relationship between the canopy color parameters and thethree growth properties at different times during the growing season. The results indicated that the temporal changes in the relationship between canopy color and thegrowth properties were similar to the temporal changes in the relationship between CCand the growth properties. Both relationships could be fitted with an exponentialfunction: y=kebx. The models were tested and the results showed that the predictionsusing G-R and2g-r-b were more accurate for LAI than for either total N content oraboveground biomass. Predictions using G/R were more accurate for abovegroundbiomass than for either LAI or total N content.4. A model to simulate the aboveground biomass of cotton based on the product ofthermal effectiveness and photosynthetically active radiation (TEP).Using computer vision technology, we developed an improved model fordescribing the effect of three factors (i.e., the spatial distribution of the cotton canopy,photosynthetically active radiation, and canopy temperature) on the accumulation ofaboveground cotton biomass. We used the abbreviation TEP to stand for the product ofthermal effectiveness and photosynthetically active radiation. The TEP value wasnormalized to obtain relative TEP (RTEP), which was then used as an input in ourmodel. The model was calibrated using data from field plots with five N rates and twocotton cultivars. Model validation was conducted using data from three independentcotton fields. Eight nonlinear functions described cotton growth well (R>0.0.894,SD<0.05). The parameters of the functions were then compared. The results indicatedthat the Richards function best fit the nonlinear relationships in a biologicallymeaningful way. The equation was as follows: relative aboveground biomassaccumulation (RAGBA)=1.024/(1+e6.646-10.115RTEP)1/1.417(r=0.981, s=0.043).Validation results indicated that the root mean square error (RMSE) was0.659t hm-2,the relative error (RE) was5.337%, the coefficient of concordance (COC) was0.988,and the coefficient of determination (R2) was0.961. The second derivative of theoptimized model showed that in cotton, the process of aboveground biomassaccumulation could be divided into three phases using two inflection points. When theaccumulation rate of the aboveground biomass of cotton was at its maximum, the RTEPwas0.622, the maximum rate of aboveground biomass accumulation was2.299, and theaboveground biomass accumulation was0.549. In conclusion, our study indicates thatTEP is a valuable parameter for estimating aboveground biomass accumulation incotton.5. A model to simulate the leaf area index (LAI) of a cotton canopy based on theproduct of thermal effectiveness and photosynthetically active radiation (TEP).Plot experiments were conducted to investigate the relationship between LAI andTEP in cotton. The study included two cotton cultivars (Shiza2and Xinluzao43) andfour N rates. We analyzed temporal changes in LAI and TEP during the entire growing season and then used Curve Expert4.1software or Origin8.5software to simulaterelative LAI (RLAI) and relative TEP (RTEP). Seven analog models gave gooddescriptions of the relationship between RLAI and RTEP. Among these models, therational function model had good biological significance and a high correlationcoefficient (r=0.9959). We concluded that among these seven models, the rationalfunction gave the most accurate description of the dynamics of LAI. We validated themodel in fields using either three high-yield cotton cultivars grown under one N rate ortwo cotton cultivars grown under five N rates. The verification results showed that theconfidence α ranged between0.0771and0.1706, the determination coefficient (R2)ranged between0.9477and0.9708, the consistency coefficient (COC) ranged between0.9867and0.9891, the relative error (RE) ranged between4.3709%and7.5403%, andthe root mean square error (RMSE) ranged between0.1425and0.2267. The resultsindicated that with RTEP as an input, the model can accurately describe temporalchanges in RLAI in all treatments. Further analysis indicated that N rate had significanteffects on LAI, especially on mean LAI, maximum LAI, and the ratio of mean LAI tomaximum LAI. We conclude that mean LAI, maximum LAI, and the ratio of mean LAIto maximum LAI could be important indicators for improving the photosyntheticcharacteristics of cotton leaves as well as cotton production.6. A remote service system platform for monitoring cotton growth and N nutrientstatus based on computer vision technology.We used computer vision technology and a CCD camera to develop an initialversion of a remote service system platform for monitoring cotton growth and Nnutrient status. The platform involves real-time monitoring technology, digital imagerecognition segmentation processing technology, agricultural internet of thingstechnology, Web network information transmission service technology, and remotedatabase management technology. The clients may use either PCs or smart phones(Android system). The remote terminal uses B/S structure. The platform provides userswith access to the Cotton Growth Monitoring Center (field monitoring), the NetworkInformation Service Control Center (server), the Image Analysis and Data ProcessingCenter, the Diagnostic Decision-making and Evaluation Center, and the User BrowsingCenter. Based on computer vision technology, this "one network, three server layers,and five centers" system can be used to remotely monitor cotton growth and N status.