基于BP-MC网络的张家港市耕地安全定量化分析  

Quantitative analysis of cultivated land safety in Zhangjiagang City based on BP-MC networks

作　　者：钱育蓉[1,2] 李建龙[1] 王卫源[3] 杨峰[1] 杨齐[1] 张洁[1] 

机构地区：[1]南京大学生命科学学院,南京210093 [2]新疆大学软件学院,乌鲁木齐830046 [3]新疆军区司令部,乌鲁木齐830042

出　　处：《农业工程学报》2009年第12期299-305,共7页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家高技术(863计划)专题项目(2007AA10Z231)的资助

摘　　要：近年来城市用地大量侵占农业用地,不断威胁到城市的粮食安全问题,利用遥感快速准确地监测土地利用覆盖变化和城市扩张对保护耕地意义重大。运用BP人工神经网络,结合广义蒙特卡罗思想,利用多年的张家港市TM遥感影像、土地利用图和年鉴,对耕地、道路、城市扩张、GDP及产业经济结构之间的关系进行了深入的分析。研究结果表明:张家港市10 a来耕地面积以每年约2.5%的幅度减少,但在2002年后农业总产值有小幅回升,这与该市落实土地优化政策、农业结构和产业结构调整息息相关。城市扩张主要方式为多核心蔓延式扩展,距主要道路和已建城区2 km内的农业用地最有可能在4 a内转变为城镇用地,并且道路和城区附近500 m内是转化高峰,而且间隔年限越长,道路和已建城区对其缓冲区的影响范围越广、辐射强度越大。经济的快速发展使得耕地大量被占用,导致了农转非概率升高,但随着城市用地与耕地数量逐渐达到平衡,农转非概率降低,使得GDP与非农转化率呈现倒"U"曲线;BP-MC网络通过在图像中采集大量像素点进行训练,有效地避免了人工神经网络陷入局部极小点,是一种定量研究城市扩张驱动力的有效途径。In recent years,cultivated land was occupied continually,which had threatened the grain safety of city. Remote sensing is a powerful tool for protecting cultivated land by means of accurate inspecting the urban expansion and land use cover and change in time.Combined with the principle of general Monte Carlo,the relationships among cultivated land,road,urban expansion,GDP and industry economic structure were analyzed deeply using BP artificial neural networks based on TM remote sensing images,land use map and yearbook of Zhangjiagang City in several years. The results demonstrated that the cultivated land area decreased about 2.5%per year in the past ten years,however the general agriculture output had a small rebound after 2002,this attributed to the government optimized policy to land use and restructure of agriculture and industry.The expansion trends of Zhangjiagang City was the spread of multi-core along the road.Cultivated land within 2 km far from the main road and city zone was most likely changed into the urban area,especially within 500 meters.The rapid development of economic resulted in an increased transformation probability of cultivated land to urban land.However,transformation probability decreased gradually with the balance between urban land and cultivated land.So the relationship between transformation probability and GDP present an inverted＂U＂curve.In conclusion,BP-MC networks deal with training set of a large number of pixels,this avoids artificial neural network stepping into the local minimum point,and it is an effective method to analyze the drive factors of urban expansion quantitatively.

关 键 词：反向传播 神经网络 土地利用 蒙特卡罗 城市扩张 耕地减少 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]