A Comparison of Sar Filtering Techniques on Agricultural Area Identification

This study presents a comprehensive evaluation of the most frequently used non-adaptive and adaptive Synthetic Aperture Radar (SAR) filtering techniques called; Mean, Median, Lee, Lee-sigma, Local Region, Frost and GammaMAP. Envisat ASAR Precision Image (PI) mode data acquired on August 2008 is used to examine the filtering techniques. Three test sites (~ 4 km), located in Karacabey of Bursa in northwest of Turkey are selected. Two of them consist of homogenous agricultural fields and the third one is selected from lake. One of the agricultural test sites has 143 fields where seven different crop types namely; corn, pasture, pepper, sugar beet, tomato, wheat, and watermelon exist. The other agricultural test site contains relatively smaller agricultural fields and it has 386 fields where corn, rice, sugar beet, tomato, wheat, and watermelon are cultivated in it. After correcting the images geometrically, the filtering operations are applied on the amplitude data using 5x5 windows. The filtering performances are evaluated by computing difference of means (MeanDif), difference of standard deviation (StdDif), correlation, and quality factor (Q). If the conditions of these evaluation indicators provide the smallest possible means, standard deviation and the maximum correlation and Q, it can be stated that the filtered image preserves the spectral information of the original image while reducing the speckle effect. Based on this evaluation the most reliable outputs are achieved by applying the Lee filter when compared with the original data. This technique is followed by the results of the Mean, Median, Gamma-MAP and Frost filters.