水稻种植面积监测是当前农业土地变化科学的热点问题,但运用遥感技术对水稻种植面积精确实施监测一直是难点。中分辨率遥感影像能够满足我国大面积水稻作物监测,成为业务化运行的主要数据源。为此,该研究尝试以中分辨率TM影像为数据源,结合神经网络和面向对象(SVM)两种算法对对黑龙江省富锦市2010年两期不同时相影像分别进行水稻分类提取,并对分类结果进行滤波处理及混淆矩阵精度评定。结果表明:(1)在高纬度单季稻生长区,面向对象分类算法的精度显著高于神经网络的分类精度,水稻用户精度和生产者精度在6月份分别高0.55%、1.37%,在8月份分别高0.62%、2.34%;(2)对神经网络分类的结果进行Majority滤波处理,在一定程度上可以改善水稻分类的精度,水稻用户精度和生产者精度在6月份分别提高0.14%、0.5%,在8月份分别提高1.56%、1.43%;(3)选取关键水稻物候期的遥感影像获取水稻种植面积的精度更高,返青期水稻提取精度要高于乳熟期,其中神经网络算法的水稻用户精度及生产者精度分别提高2.67%、3.45%;面向对象算法的水稻用户精度及生产者精度分别提高2.6%、2.48%。未来需要重点考虑建立全国水稻物候历信息、面向对象算法中自动化最优尺度分割方法来提高水稻分类的精度。
Monitoring the rice area has become a hot issue of the present agricultural land change study. However, itg uneasy to accurately monitor rice area using remote sensing technology. Based on two phase mid - spatial resolu- tion TM images, this paper monitored the rice area in Fujin County, Heilongjiang Province using the methods of ob- ject oriented classification and Artificial Neural Network (ANN). The conclusions were as follows: Firstly, the ac- curacy using the method of object - oriented classification was higher than that using ANN classification method. It was 0. 55% and 1.37% higher for rice user and producer accuracy on June, and 0. 62% and 2. 34% on August. Secondly, after the process of Majority analysis, the rice classification accuracy would increase by 0. 14% and 0. 5% on June, and 1.56% and 1.43% on August for rice user and producer accuracy using the method of ANN. Thirdly, choosing the proper time images can get a higher accuracy rice area, and the accuracy obtained from retur- ning green stage was better than that from the milk ripe stage, the rice user accuracy and producer accuracy can in- crease 2. 67% and 3.45% using ANN method and can increase 2. 6% and 2. 48% using the method of object-ori- ented classification. In future, it should improve the classification accuracy through building long time series of rice phonological calendar and using the method of object-oriented automatized segmentation scale.