基于自组织神经网络的耕地自然质量评价方法及其应用

Method and its application of natural quality evaluation of arable land based on self-organizing feature map neural network

耕地质量各构成要素的特点和相互间的影响,决定了耕地质量的外在表现,客观地确定耕地自然质量,对耕地分等定级具有重要的意义。该文通过对已有耕地质量评价方法的优势与不足的分析,提出在空间数据库基础上应用自组织神经网络的耕地自然质量评价方法,并应用该方法对吉林省九台市耕地自然质量进行了评价,通过步长为1 000次训练,自动生成13个类别,在13个类别基础上按照九台市的指定作物的光温生产潜力指数、作物的产量比系数,进行了耕地自然质量评价。根据评价分值的大小分为3等,其中质量等级Ⅰ级占全市耕地面积42.13%,Ⅱ级占全市耕地面积30.40%,Ⅲ级占全市耕地面积27.47%。评价结果与《九台市耕地质量更新成果》比较,图斑重合率为80.78%,面积重合率为79.42%。2种评价方法可能出现差异的原因:该文评价方法增加了坡度因子,且《九台市耕地质量更新成果》采用的是全省统一的指标权重;2种方法对于一些定性描述指标均通过信息赋权值法进行量化,而2种方法中量化方法不同,赋值不同。该方法将自组织神经网络和地理信息系统相结合,有效地集成影响耕地质量相关的土壤及土壤环境信息,利用自组织神经网络在没有教师信号时自动连接权值向着更利于竞争方向调整,通过度量评价单元的相似程度,使类间差异最大而类内差异最小,逐步将评价单元划分类别。根据每个类别中图斑自然质量指数的大小进行耕地质量等别评价,提高了评价结果的可信度,为耕地质量评价提供了新思路。

The characteristics and interactions of arable land quality components determine the external manifestation of arable land quality. It was the vital significance to objectively determine natural quality of arable land for arable land classification and grading. In this paper, advantage and disadvantage of the existing methods used in the arable land natural quality evaluation was analyzed, SOFM（self organizing feature map） neural network method was proposed to evaluate arable land natural quality basing on spatial database in Jiutai city of Jilin province. The nine evaluation indicators including surface soil texture, profile configuration, content of soil organic matter, soil p H value, barrier layer depth, soil salinity, effective soil depth, drainage condition and slope, were chosen and the corresponding database layer was established in the method. At the same time, attribute values were input and data were normalized. By training step for 1000, the system automatically generated 13 categories. Based on temperature potential productivity index and crop yield ratio of Jiutai city, evaluation index of arable land natural quality were calculated based on GIS. According the size of arable land natural quality indicators, the arable land natural quality of Jiutai city was divided into 3 grades. The ratios of Ⅰ, Ⅱ and Ⅲ grade of arable land nature quality which was classified with proposed method in this paper to the total arable land area in Jiutai city were 42.13%, 30.40%, 27.47%, respectively. The natural quality evaluation results of arable land based on SOFM neural network were compared with that with farmland natural quality grading of update results in Jiutai city. The comparison results indicated that in the nature quality grade Ⅰwith proposed method in this paper, the ratio of fifth, sixth and seventh grade which was classified with farmland natural quality grading of update results was 33.91%, 4.25%, 3.98%, respectively; in the nature quality grade Ⅱ with proposed method in this paper, the ratio of fifth, sixth and seventh grade which was classified with farmland natural quality grading of update results was 4.58%, 25.38%, 0.45%, respectively; in the nature quality grade Ⅲ with proposed method in this paper, the ratio of fifth, sixth and seventh grade which was classified with farmland natural quality grading of update results was 0.85%, 6.48%, 20.13%, respectively Quality grade Ⅰ, Ⅱ and Ⅲ, classified with SOFM neural network corresponded to fifth, sixth and seventh grade of national nature grades classified with farmland natural quality grading of update results, figure spot coincidence rate and area overlap rate were 80.78% and 79.42%, respectively. The seasons for the discrepancies in the results of the two kinds evaluation method were that the slope factor was considered in proposed method, at the same time, uniform weights in Jiutai city, Jilin province was used in arable land natural quality grading of update results and qualitative description of the two kinds of evaluation methods for some indicators were quantified through information weighting value method, due to different quantization methods, different assignments. In this method, by self-organizing neural network combining with geographic information systems based on MATLAB, effectively influencing factors of arable land quality that related to soil and soil environment were integrated and quantification and space for spatial and non-spatial datum was realized. Especially, weights were subjectively determined, which made evaluation results more objective and true in propose method. And this method had more advantages in evaluating for highly nonlinear relationship between the arable land quality and its affecting parameters. Overall, the study method provides a new idea for arable land quality evaluation and broadens depth and breadth of arable land quality evaluation, and enhances the results credibility of natural quality evaluation of arable land.