| Relying on Research and Development as well as Application of Precision Agricultural Machinery Operation Collaborative System(No.2013AA10230803),a research project of the National High-tech R&D Program(863 Program),and Research and Application of Grid-based Land Management and Service Platform in Villages and Small Towns,a research project of National Surveying and Mapping Geographic Information Bureau,with the Yanzhou District of Jining City and Xanthoceras sorbifolium bunge test bases in Linzi District of Zibo City,Shandong Province as the study area,this study was carried to explore multi-perspective grid division of farmland and its application from three different perspectives,i.e.macroscopic(administrative area),mesoscopic(farming area)and microscopic(single crop).The research content and conclusions of this study are as follows:(1)Grid division of farmland in precision agriculture was studied,and the methods of grid division were constructed from different perspectives.From the macroscopic perspective,according to the experience of grid division of urban communities,the principle and scheme regarding grid division of farmland were determined.From the mesoscopic perspective,the optimal grid size of farmland was explored based on the current grid division of farmland in Yanzhou District,concluding 400 mu as the grid size most appropriate for Yanzhou District.From the microscopic perspective,it was determined through experiments that the suitable grid size for Xanthoceras sorbifolium bunge management was 3 m ? 4 m.(2)From the macroscopic perspective,this study started with social management and services to probe into farmland management in precision agriculture,constructed a grid-based farmland management model,verified the rationality of the management model,and quantitatively compared grid and non-grid farmland management models.Based on the experience accumulated in grid management of digital cities and the current grid management of rural communities in Yanzhou District,the existing farmland management process was reengineered and optimized,and a universal farmland management process featuring “seven-step closed-loop business collaboration method” was constructed.A structural verification method combining Petri net reduction technology and logical table was proposed,in the hope of avoiding structural errors in the model and theoretically guaranteeing the successful implementation of the management model in the later stage.A PIPE simulation experiment was conducted,the results of which further verified the rationality of the farmland grid management model and the effectiveness of the proposed verification method.The grid and non-grid farmland management models were quantitatively and objectively compared using the Arena simulation tool and the quantitative measurement model established based on stochastic Petri net.According to the comparison results,the grid farmland management model has significant advantages over the non-grid one,providing a scientific basis for determining whether the grid farmland management model can be implemented and promoted in the later period.(3)From the mesoscopic perspective,this study explored how to determine reasonable soil sampling points based on gridding and verify the rationality,obtaining the optimized sampling points for many years.There were 86 sampling points with an interval of about 1.5 km in 4 towns in Yanzhou District,i.e.20 in Xiaomeng Town,20 in Caohe Town,28 in Da'an Town and 18 in Xinyan Town,68 of which were located in the Moisture cinnamon soil area,12 located in the Mortar black soil area and 6 in the Moisture soil area.The rationality of the determined soil sampling points was comprehensively verified from different angles and emphases.The basic descriptive statistical results show that although the number of sampling points was reduced,yet the average,median,variation coefficient,variation degree and standard deviation of each soil nutrient were still quite close to the statistical results gained based on the original sampling data.According to the classic Cochran formula,86 sampling points are completely able to meet the requirements of measuring the soil fertility.The results of geostatistical analysis imply that the variation range of each soil nutrient is greater than the sampling interval.The ordinary Kriging interpolation method was used for spatial estimation,and the interpolation accuracy was evaluated through cross-validation.According to the results,the mean error(ME)and the mean standard error(MSE)are small,both of which are close to 0;the root mean square standard error(RMSSE)is approaching 1;the root mean square error(RMSE)is very close to the average standard error(ASE).On this basis,the soil sampling scheme that had been proven rational was applied to optimize the existing soil sampling,obtaining the the sampling points data for many years which number and distribution were basically consistent.(4)Based on the rational grid soil sampling data,a model of predicting the variation trend of soil fertility was established.From a socioeconomic perspective,based on the rational grid soil sampling data from 2012 to 2017,the influencing factors of soil fertility were analyzed.According to the model of predicting the variation trend of soil fertility established based on stochastic Petri net,there is a probability of about 0.7852 for soil fertility in the study area to decline in the next year.By comparing the soil fertility in 2016 and 2017,and further analyzing the variation of soil fertility from 2012 to 2016,the proposed method for predicting changes in soil fertility was verified to be effective.(5)Based on the rational grid soil sampling data,the spatial variation of soil nutrient was studied,and a query “map” of spatiotemporal variation of soil nutrient and fertility was made.First of all,a descriptive statistical analysis was performed on soil nutrients based on the rational grid soil sampling data rom 2012 to 2017.According to the analysis results,the content of rapidly available potassium,alkaline hydrolysis nitrogen,and organic matter were on a downward curve from 2012 to 2017;the content of available phosphorus had been at level 2.Soil pH is weakly variable,while available phosphorus,organic matter,rapidly available potassium and alkaline hydrolysis nitrogen are all moderately variable.Based on the divided farmland grid and the soil data in 2017,the spatial variation of soil nutrients in the study area was explored,the results of which show that the content of organic matter was basically distributed in stripes,with the content gradually decreasing from the west to the east.The content of organic matter in all the farmland grids was at the lower-middle level 4.On the whole,the content of alkaline hydrolysis nitrogen is higher in the north and south parts,and lower in the north central part.However,the content of alkaline hydrolysis nitrogen in most farmland grids was at the medium level 3.The content of available phosphorus gradually decreased from the north to the south,and the content of available phosphorus in most farmland grids was at the above-average level 2.As a whole,the content of rapidly available potassium gradually declined from the southwest to the northeast,with the content of rapidly available potassium in most farmland grids at the medium level 3.The content of all the 4 soil nutrients was relatively high in Xiaomeng Town.On the basis of the above research,a query “map” of spatiotemporal variation of soil nutrient and fertility provides an access to conditions about soil nutrient and fertility in the overall study area as well as on every single grid,supporting the decision-making of precise fertilization.(6)From the microscopic perspective,the problem of accurate harvest of field Xanthoceras sorbifolium bunge based on grid recognition was studied,a Xanthoceras sorbifolium bunge image collection system was designed and a deep learning model for recognizing mature Xanthoceras sorbifolium bunge was established.A Xanthoceras sorbifolium bunge image collection system was researched and developed,achieving functions such as field walking,grid recognition as well as the acquisition,transmission and storage of data.Moreover,grid recognition was explored as well,and the system test results show that the location and number of specific grids can be obtained automatically based on the GPS coordinates.In addition,this paper probed into the recognition of mature Xanthoceras sorbifolium bunge,and established a deep learning model for recognizing mature Xanthoceras sorbifolium bunge.According to the test results,in the original data set,the trained optimal model can accurately recognize 81% mature and 82% immature Xanthoceras sorbifolium bunge respectively.The recognition model was evaluated with 4 indicators,i.e.the accuracy rate,precision rate,recall rate and F1_Score of recognition.Based on the evaluation results,each indicator value of the optimal model can reach 80% in either the original data set or the simulated data set.It indicates that the constructed model for recognizing mature Xanthoceras sorbifolium bunge can serve as a tool to identify the ripeness of Xanthoceras sorbifolium bunge.By comparing this model with models that do not use simulated data,it is found that the data simulation technology is able to expand the training data set,thereby improving the generalization of the model and the accuracy of prediction as well as effectively addressing the problem of “overfitting”. |
PDF Full Text Request