An approach for the automatic segmentation of high resolution satellite images into agricultural fields

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Hacettepe Üniversitesi, Fen Bilimleri Enstitüsü, Geomatik Mühendisliği A.B.D., Türkiye

Tezin Onay Tarihi: 2015

Tezin Dili: İngilizce

Öğrenci: ALIREZA RAHIMZADEGANASL

Danışman: Mustafa Türker

Özet:

Monitoring and management of land use plays an important role in the economic development of agriculture regions in the world. Satellite images have become very popular data source for crop mapping in agricultural areas. With this respect, the extraction of agricultural land parcels is critical for crop classification and further field-based analysis operations. An efficient image analysis procedure is needed to automatically extract agricultural fields with the minimum user intervention. When performed on field-by-field basis, agricultural crop classification produces much better results. Field-based image classification techniques use approaches that assign crop labels for the agricultural fields individually. The classification is performed within permanent field boundaries which are stored in a geographic information system (GIS) database as vector dataset.

In this study, a field-based segmentation approach is proposed to extract sub-fields within permanent boundaries of agricultural fields from high resolution remotely sensed imagery. The sub-field extraction operation is carried out one field at a time by means of processing each field separately. The developed approach is based on the extension of the watershed algorithm for multispectral image segmentation. Watershed transformation is a powerful algorithm for image segmentation. First, a gradient magnitude image is calculated from the generated intensity image using the Sobel edge detection operator. Next, the foreground markers, which will be used in watershed transformation, are extracted using Otsu’s thresholding method. To automate the marker extraction procedure, Otsu’s thresholding method is employed in an optimized iterative manner. The extracted foreground markers are used in the remainder of the process after performing some morphological operations. A marker controlled watershed segmentation is then carried out for the extraction of sub-fields that stay within permanent boundaries of agricultural fields.

The developed approach was tested in an agricultural area locate near the town of Karacabey, Bursa in North-West of Turkey. The high resolution satellite images used include the Spot5 multispectral (XS) image acquired in 22 July 2004, the Ikonos multispectral (XS) image acquired in 15 July 2004 and the QuickBird multispectral (XS) and pansharpened (PS) images acquired in 13 August 2004. The achieved results are quite promising. The overall accuracies of the sub-fields extracted through developed automatic approach were computed to be 80.34%, 89.72%, 83.24% and 78.88% for the Spot5 (XS), Ikonos (XS), QuickBird (XS) and QuickBird (PS) images, respectively.