고해상도 위성영상은 지표면의 모습을 자세하게 담고 있기 때문에 상세한 지리정보를 추출할 수 있다. 최근 고해상도 위성영상이 활발히 이용되면서, 새로운 영상분류 기법이 모색되고 있다. 본 연구에서는 영상에서 추출된 객체를 대상으로 분광 정보와 텍스쳐 정보를 이용한 새로운 분류 기법을 개발하였다. 훈련지역의 다양한 정보를 기계 학습 기법에 적용하고, 그로부터 도출된 규칙을 이용하여 규칙 기반 분류기법에 적용하였다. 사례 지역의 자료를 이용하여 기계 학습 기법과 규칙 기반 분류에 적용한 결과, 기존의 분광 특성만을 이용한 영상보다 분류 결과의 정확도 측면에서 우수한 것으로 나타났다. 기계 학습 분류 기법과 규칙 기반 분류 기법은 분광 특성뿐만 아니라 다양한 텍스쳐 정보를 함께 고려하기 때문에 보다 간편하고 효과적인 영상분류 기법이라고 할 수 있다.

The high resolution satellite image can provide more specific geographic information since it contains the details of the earth surface. As very precise images including high resolution images are more available, we have tried to have more effective techniques for the digital image processing. This research proposes to use the spectral and texture information to develop a new classification technique that can be applied to the object extracted from satellite image. The researcher set up a classification rule by taking the machine learning approach to the attribute data of the training set. And the rule is further used to classify the satellite image. A test with case study data shows this rulebased classification technique using machine learning approach could be useful since the approach allows us to consider the various spectral and textual information and the rule-based classification is simple and more effective than traditional classification.

The high resolution satellite image can provide more specific geographic information since it contains the details of the earth surface. As very precise images including high resolution images are more available, we have tried to have more effective techniques for the digital image processing. This research proposes to use the spectral and texture information to develop a new classification technique that can be applied to the object extracted from satellite image. The researcher set up a classification rule by taking the machine learning approach to the attribute data of the training set. And the rule is further used to classify the satellite image. A test with case study data shows this rulebased classification technique using machine learning approach could be useful since the approach allows us to consider the various spectral and textual information and the rule-based classification is simple and more effective than traditional classification.

machine learning approach, object-based classification, rule-based classification, high resolution satellite image