Appraisal Of Cultivated Land Quality Grade Based On BP Neural Network And National Standard Method

To promote rural revitalization and speed up building a strong agricultural nation,our Party and country’s three rural themes are our highest priorities,and farmland is a solid foundation for agricultural development.The traditional methods of evaluating the quality of farmland are less efficient and more subjective.And the existing process of dividing the quality level of arable land(GBT/33469-2016)for the presence of contaminated arable land directly in the arable land soil pollution evaluation process,not involved in the evaluation of the quality of arable land.There is a lack of arable land quality classification process and related evaluation for such arable land with soil contamination level between screening and control values.This is not in accordance with the actual situation,and the problem of crops,vegetables and exported agricultural products grown on arable land in the safe use class of classification still exists.In addition,the machine learning artificial neural network method does not need to determine the weights,avoids the influence of human components in the traditional evaluation process,has better intelligence,and can cover the heavy metal pollution factors in agricultural soils,so as to objectively obtain the quality level of farmland information.Therefore,it is important to carry out arable land quality evaluation by artificial neural network algorithm to map the status of arable land resources and implement arable land quality improvement change,so as to improve food production,which is important for land reclamation,ensuring food security and accelerating the construction of a strong agricultural country.This paper takes Tongling City,Anhui Province as the research object,and constructs evaluation models using artificial neural network algorithm and national standard method,which is Telfer method,hierarchical analysis method and fuzzy evaluation method,respectively.By comparing the evaluation results of the two methods in terms of area share and spatial distribution,the grade classification accuracy of the BP neural network model of farmland quality evaluation results is 98% with the evaluation results of the hierarchical analysis method as the benchmark,and finally the sensitivity analysis ranking of the influencing factors of farmland quality grade is calculated based on the BP neural network model.With the following main research results in this paper:(1)From the screening,all 15 assessment indicators were highly correlated with the level of soil quality.The correlations were ranked as potential ecological risk > capacity >organic matter content > effective soil thickness > tillage texture > topographic location >constraint factors > fast acting potassium > irrigation capacity > biodiversity > farmland forestry network > texture configuration > effective phosphorus > p H > drainage capacity.All 15 selected evaluation indices can be used in the construction of the BP neural network farmland quality level evaluation model.(2)The BP neural network model for farmland quality assessment in Tongling was established,and its overall accuracy R reached about 0.98.The difference between the predicted quality of its test samples and the actual quality error is also small.This result shows that the BP neural network method is feasible in the field of farmland quality evaluation,and its prediction accuracy is also high and can meet the requirements of farmland quality evaluation.(3)The results of the BP neural network model and the evaluation model of the national standard method were compared.There is only an error of 0.04 in the quality rating of farmland in Tongling between the two,and the evaluation results based on the BP neural network model are generally consistent in terms of area and percentage distribution of farmland quality rating compared with the results of the national standard method,with only slight differences in spatial distribution.(4)Based on the evaluation results of the BP neural network model in Tongling,the sensitivity of the factors influencing the quality level of arable land was analysed.The ranking was as follows: potassium content > p H > texture of cultivated layer > texture configuration > effective phosphorus content > capacity > effective soil thickness >organic matter content > biodiversity > drainage capacity > topographic parts > soil pollution status > irrigation capacity > obstacle factors > degree of farmland forestry network.(5)As proved,the BP neural network model can avoid subjective weighting errors and has good evaluation efficiency,which can meet the accuracy requirements of arable land quality evaluation and can be applied to arable land quality evaluation operations.This study can provide a reference for the application of new technology of arable land quality grade evaluation and a basis for the improvement of arable land quality construction.