بهبود طبقه بندي طيفي- مكاني جنگل پوشاي مينيمم با كاهش ابعاد تصاوير فراطيفي

Improved Spectral-Spatial Classification Minimum Spanning Forest by Reducing the Spatial Dimensions of Hyperspectral Images

فن‌آوری سنجش از دور فراطیفی دارای كاربردهای فراوان در طبقه‌بندی پوشش‌های زمین و بررسی تغییرات آنها است. با پیشرفت‌های اخیر و ایجاد تصاویری با قدرت تفكیك مكانی بالا، لزوم استفاده توام از اطلاعات طیفی و مكانی را در طبقه‌بندی تصاویر فراطیفی ایجاب می‌كند. در این تحقیق روشی جدید جهت طبقه‌بندی طیفی-مكانی تصاویر فراطیفی به كمك الگوریتم جنگل پوشای مینیمم ( MSF) مبتنی بر نشانه‌ها كه یكی از دقیق‌ترین الگوریتم‌ها در این زمینه است و تكنیك كاهش ابعاد معرفی می‌شود. در روش پیشنهادی تاثیر كاهش ابعاد تصاویر فراطیفی به كمك الگوریتم ژنتیك در سه مرحله قبل و بعد از انتخاب نشانه‌ها و به صورت همزمان بررسی می‌گردد. در این مطالعه نشانه‌ها از روی نقشه طبقه‌بندی ماشین بردار پشتیبان ( SVM) انتخاب شدند. روش‌ پیشنهادی بر روی سه تصویر فراطیفی Pavia، Telops و Indian Pines پیاده‌سازی گردید، نتایج آزمایشات بدست‌آمده برتری به كارگیری الگوریتم ژنتیك را قبل از انتخاب نشانه‌ها در تصاویر Pavia و Telops نشان می‌دهد. در تصویر Indian Pines كاهش ابعاد در هر دو مرحله قبل و بعد از انتخاب نشانه‌ها و به صورت همزمان موجب افزایش دقت طبقه‌بندی می‌گردد.

Imaging spectroscopy, also known as hyperspectral imaging, is concerned with the measurement, analysis, and interpretation of spectra acquired from either a given scene or a specific object at a short, medium, or long distance by a satellite sensor over the visible to infrared and sometime thermal spectral regions. The recent developments in spatial, spectral and radiometric resolution of hyperspectral images have stimulated new methodologies for land cover and land use classification. There are two major approaches for classification of hyperspectral images:  the spectral or pixel-based and the spectral-spatial or object-based approaches. While the pixel-based techniques use only the spectral information of the pixels, the spectral-spatial frameworks employ both spectral characteristics and spatial context of the pixels. The pixel-based classification methods are often unable to accurately differentiate between some classes with high spectral similarity. This is mainly because they employ only the spectral information in order to identify different land covers. Consequently, methods that can exploit the spatial information are essential for more accurate classification results. Among the various methods for extracting spatial information, segmentation techniques are the powerful tools for defining the spatial dependences among the pixels and finding the homogeneous regions in the image. An alternative way to achieve the accurate segmentations of image is marker-controlled segmentation. The idea behind this approach is selecting of one or several pixels for every spatial object as the seed or a marker of the corresponding region. The marker-based segmentation significantly reduced the over-segmentation problem and led to better accuracy rate. Recently, an effective approach to spectral-spatial classification of hyperspectral images has been proposed based on Minimum Spanning Forest (MSF) grown from automatically selected markers using Support Vector Machines (SVM) classification. In this framework, a connected components labelling is applied on the classification map. Then, if a region is large enough, its marker is determined as the P% of pixels within this region with the highest probability estimates. Otherwise, it should lead to a marker only if it is very reliable. A potential marker is formed by pixels with estimated probability higher than a defined threshold. This paper aims at improving this approach by reducing the spatial dimensions of hyperspectral images. The proposed approach are evaluated the dimension reduction of hyperspectral image before and after marker selection process in MSF using genetic algorithm. The genetic algorithm is a general adaptive optimization search method based on a direct analogy to Darwinian natural selection and genetics in biological systems. It starts from an initial population which is composed of a set of possible solutions called individuals (chromosomes), and then evaluates the quality of each individual based on a fitness function. We use the Kappa coefficient accuracy parameter of SVM classification obtained from the training samples subset as the fitness function. Three benchmark hyperspectral datasets are used for evaluation:  the Pavia dataset, the Telops dataset and the Indian Pines dataset. Experimental results show the superiority of using genetic algorithm before selecting markers in Pavia and Telops datasets. In Indian Pines dataset, the classification accuracy was increased with reduced dimensions both before and after the marker selection and concurrently.