Study On Non-Destructive Monitoring Of Sugarcane Growth In On-Line Image

Crop growth monitoring is an important component of Agro-meteorological observations as it can provide basic data for crop cultivation field management and yield forecasting.Traditionally,crop growth monitoring is mainly carried out by field observation,which is subjective,time-consuming and labor-extensive.With the extensive application of computer technology and the Internet in agriculture,artificial observation methods cannot meet the needs of modern agriculture development.Therefore,an on-line,real-time,and automated observation is needed to monitor crop growth.In the field environment,the use of imaging equipment combined with computer,image processing,internet technology for real-time monitoring of crops is an alternative method.In this paper,field experiments were carried out in sugarcane fields in Shechong Township,Liuzhou City,Guangxi Zhuang Autonomous Region,and the sugarcane images were extracted and analyzed by using the all-factor agricultural automatic meteorological observation system in order to provide technical basis for Sugarcane growth on-line quantitative non-destructive monitoring.Sugarcane photo images were obtained by the automatic image acquisition device(CCD video camera and digital camera)during the whole growth period in the year of 2015 and 2016.These images were analyzed and processed based on image processing technology.Color indices generated by RGB component,sugarcane leaf area index simulating model and sugarcane coverage estimating model were developed to reflect the growth situation.The automatic detection of sugarcane emergence was carried out using image segmentation,image binarization,connected domain marker,connected domain feature statistics,etc.The results of this study show that:(1)Sugarcane leaf area index simulating model was developed based on color indices,which is composited by RGB bands.G-B color index is found to be the optimal indicator in estimating LAI of sugarcane and the estimated accuracy of LAI model is highest when the G-B color index is the average value at 9:00AM,11:00AM and 3:00PM.The maximum predicted R2 and minimum RMSE is 0.8164 and 0.1211,respectively.(2)The reference fraction vegetation cover(FVC)of sugarcane was calculated using the PFMRF method.Then nine color indices of the images were calculated to analyze their correlation with reference FVC.The results show a good correlation between reference FVC and the color indices.The accuracy of sugarcane FVC estimating model based on the indices of ExG-ExR,ExG and CIVE are all better.Obviously,the optimum color index is ExG,which is estimated by using the image shooting at 11:00 Am.The accuracy of ExG-based model is the highest as well with its RMSE and MAE of 0.0484 and 0.0409,respectively.(1)The weeds are successfully removed and sugarcane seedling is accurately captured by using the shape of the sugarcane seedling on the image.The distribution characteristics of the identified sugarcane seedlings were analyzed,and then the image was judged to reach the seedling stage.Overall,this automatic identification algorithm for sugarcane seedling was rather accurate with the error within ± 3 days validating with the observed sugarcane development period.