论地理国情监测的技术支撑

针对地理国情监测中大幅面多时相遥感影像变化检测的需求，提出了一种基于卡方变换和样本选择的面向对象遥感影像变化检测方法。首先对多时相遥感影像进行多尺度分割获取像斑；然后，提取像斑的多维特征，采用基于卡方变换的特征融合方法计算像斑的加权差异度；最后，自适应选择训练样本，通过基于最大期望算法的贝叶斯阈值确定方法获取变化阈值，并对加权差异影像进行二值分割获取变化检测结果。
以武汉市东湖高新技术开发区为例，利用多时相高分辨率遥感影像进行土地覆盖变化检测。试验结果表明，该方法可以克服全样本变化向量分析法及全样本卡方变换检测法难以满足阈值确定条件的不足，获得更准确的变化阈值，保证变化检测正确率高而又有效地降低漏检率，从而获得更好的变化检测结果，在地理国情监测中具有一定的应用价值。

Discussion on Technologies of Geographic National Condition Monitoring

In this paper, an object-based change detection method for multi-temporal remote sensing images based on the Chi Square Transformation (CST) and sample selection is proposed to measure change in the large-size multi-temporal remotely sensed images used for monitoring national geographic conditions, In this new change detection method, image segmentation is used to obtain image objects. Secondly, multiple features are extracted from image objects, and a weighted difference is calculated for each image object based on CST. Then, with adaptively selected training samples a change threshold is automatically calculated using Expectation Maximization (EM) and a Bayesian rule with a minimum error rate.
The weighted difference image is segmented into a binary image with a change threshold to derive change detection results. Multi-temporal high-resolution images of the Wuhan East Lake New Technology Development Zone were used for land cover change detection, experimental results show that the proposed method can obtain the most accurate change threshold among three tested methods. These highly accuracy change detection results effectively reduce the rate of lost detection, and are currently used for monitoring geographic conditions.