Study Of The Impact Of Remote Sensing Image Classification Accuracy Based On Multi-scales

As the remote sensing technology developing, especially with the appearance of high resolution remote sensing, human's capability of obtaining abundant information about nature improves, which enlarges our scope of observing nature. Compared with the low or middle resolution image, the high resolution remote sensing image has richer structure and texture information. However, the result of traditional statistical classification technology based on pixel's spectrum has some limitation, such as precision of classification is very low, the image information couldn't be fully utilized, and processing speed is slow. Further more, it will result in not only reducing the accuracy of classification but also making the spatial data redundant and wasting the resource.The idea of object-oriented is introduced into the information extraction technology of the high resolution image in this thesis. This technology produces homogeneous image objects through segmentation technology, and then provides a way to analyze objects'features. At last it carries out the information extraction by fuzzy classification.Generally, multi-scale segmentation technique is utilized when doing image segmentation. In the multi-scale segmentation, each object layer has its own fixed-scale value, so that multiple object levels can reflect a variety of surface category properties of different spatial scale, after which, the classification information of different properties could be extracted on different object scale layers. Hence, the problem that using the same resolution image to identify all the spatial properties may introduce incorrect types can be solved.In terms of scale selection during the process of multi-scales segmentation, this paper proposed an optimal segmentation scale selection method based on scales comparison. It justifies quantitatively the segmentation results by classification precision, so that proper spatial scales for each different types of objects can be determined. Experiment has been done to prove the effectiveness of this method. The method provides an inspired idea for the evaluation of segmentation results, as there's no standard and reliable way of evaluation all over the world.On the basis of the above-mentioned theory research, taking a typical region of Wuchang District,Wuhan City as the test area, a case study on Quickbird image classification with object-oriented approach is carried out. Scale comparison method is applied to select the segmentation scales in the experiment. The results demonstrate that scale comparison is an effective optimal segmentation scale selection method. Using object-oriented image classification method to extract information from the high-resolution remote sensing image, the extracted surface features keep consistent with the real features in both shape and property, the classification precision is enhanced, and the classification results are more easy to be interpreted and understood.