地物大小、对象尺度、影像分辨率的关系分析

遥感数据的分辨率越来越高, 给地物信息提取提出了新的挑战。利用基于像元的分类技术和基于多尺度分割的面向对象分类技术对高分辨率影像进行分类实验, 分析地物大小、对象尺度与影像分辨率的关系。实验结果表明不同地物由于其空间尺度不同, 与之相适宜的空间分辨率和对象尺度也不同, 在适宜分辨率的影像提取有较高的精度, 在适宜的对象尺度上提取对象信息有更高的精度。分析也表明面向对象的多尺度影像分类技术适应了不同地物有其相适宜的空间分辨率, 在适宜尺度影像层中提取地物, 其分类精度大大高于基于像元的分类方法。

Analys is to the Relationship of Feature Size, Objects Scales, Image Resolution

Information extraction from remote sensing data encountered a new challenge while the spatial resolution is increasing quickly. People suppose that the higher the spatial resolution is, the better the result of objects classification is. To prove this guess we use two approaches: pixel-based classification and object-oriented analysis. The former site test shows one class has different precision from various resolution images. Some classes improve their precision with the high resolution but others don’t. Without various resolution image data, how can we acquire the best information with only one image? Object qriented approach offers a good solution with its key technology: multi-scale image segmentation. Object-orien ted image analysis does no t classify single p ixel bu t rather image ob ject s. No t on ly spectral information but also spatial, physical and contextual characteristics of image objects are used for classification. The site resu lt show s that ob ject s have their best scale image levels to class. The precision of object-oriented approach is much higher than those of based-pixel approach. It makes us believe that this process is the best selection for high-resolution image analysis. Two test results have proved that all objects cou ldn’t be extracted well from the same resolution or scale. They need the corresponding resolution images or scale image levels. Multi-scale image analysis is the perfect method.