Study On Rapid Yield Measurement In Field Based On 2D Image Modeling Of Rice Panicle

Rice is the main food crop in China,and its yield estimation is of great importance for monitoring crop growth,breeding high-yield cultivars,guiding agricultural production,and regulating the balance of national grain supply and demand.As an important research topic in the field of precision agriculture,yield estimation has been the subject of many reports.Although direct field yield measurement has high accuracy,it requires considerable manpower.At present,remote sensing and crop simulation models for yield estimation generally have some limitations,such as low accuracy and low real-time performance.Therefore,it is imperative to develop labour-saving,rapid and accurate field rice yield measurement technologies.A method based on 2D image modeling of rice panicle is used in this study,1198 panicles of 6 different rice cultivars was taking as the research object.Firstly,the yield parameters and morphological structural parameters of rice panicles were collected to analyze their distribution characteristics,and the main structural characteristics determining grain weight were identified by regression analysis.Secondly,the correlation model between grain area and weight parameters was established,and the validity of the model was verified by using such indicators as coefficient of determination(R2),mean relative error(MRE),mean prediction error(MPE),prediction error standard deviation(SDPE)and root mean square prediction error(RMSPE).Finally,the method of rice field rapid yield measurement was proposed,and the "five-point calibration model" was used to obtain rice yield data quickly,compared estimated values with the actual yields.On this basis,based on the Android platform,the rapid field production software for rice field was developed,and the software was tested by posterior test.The results are as follows:(1)The distribution characteristics of rice panicle weight of six different cultivars showed that the skewness coefficient and kurtosis coefficient of the overall distribution of panicle weight of each cultivars were close to 0,indicating that the panicle weight of each cultivars was normal distribution.This indicates that the samples collected are statistically significant.The variance analysis of the measured values of six different rice panicles traits showed that there were significant differences among the different traits of the six cultivars,but the overall variation of the same traits was similar.The standardized coefficients between panicle weight,filled grain weight and grain area,theoretical height,bulk density of six different rice cultivars showed that the main determinants of panicle weight and filled grain weight were grain area.(2)The image extraction algorithm of rice panicle image was developed and the complete extraction rate was used to test algorithm.The results showed that the complete extraction rate of different rice cultivars was above 90%,and the highest extraction rate was 95.5%.The image feature extraction algorithm has good accuracy and universality,it is feasible to use image processing methods to obtain the grain area characteristics of rice panicle.(3)The "weight prediction model" and the "5-point calibration model" were established.The results showed that the determination coefficient of "weight prediction model"established between the grain area and the weight parameters of the panicle is above 0.8 and the highest can reach 0.96.The use of rice grain area to predict weight is more feasible.The MPE value of the "quality prediction model" is close to zero and the model accuracy is high.The determination coefficient of the"5-point calibration model" of different cultivars is above 0.99.The "5-point calibration model" can be used to predict the weight parameters more quickly and accurately.Comparing the SDPE and RMSPE values of different models,the SDPE and RMSPE of the G4 model is lower than other models and it’s accuracy is higher.The calculation formula of yield prediction is simplified as the product of the number of panicles per unit area and the average total grain weight.The lowest relative error of estimated yield is 1.36%,the maximum is 8.64%,and the estimation error is below 10%.(4)Based on the Android platform,the rapid yield measurement software can accurately extract the grain area of rice panicle,and the complete extraction rate is above 90%,which has certain accuracy and versatility.The accuracy level of software model is 1 st-good and 2nd-qualified.The relative error between estimated yield and actual yield is below 10%.The accuracy of the rapid yiel measurement software in rice fields has reached the available requirements,and it is possible to achieve rapid yield measurement in rice fields within a certain range.The results show that the rice yield estimation model based on 2D images of rice panicle is a fast and effective method.Combined with the portability and low cost of the Android platform,it has certain reference value for improving the automation level and breeding efficiency of field measurement in China.