Texture analysis and digital classification of SAR images for agricultural target discrimination

Abstract

The remote sensing images of the visible and infrared range of the electromagnetic spectrum have great potential for discriminating agricultural areas for the purpose of estimating the yield. Nevertheless, the presence of clouds prevents the acquisition of this type of images, however, the SAR images are independent of meteorological conditions. In this context, this work verified the potential of two SAR/Radarsat-1 images, C band, HH polarization, in Fine-5/upward (F5A) and Standard-7/downward (S7D) modes in the discrimination of agricultural targets in the region of Assis-SP. The methods were based on visual analyzes and on the comparison of the digital classifications of the original and filtered F5A and S7D images, as well as their texture measurements. The results indicated that the filtered images improved the discrimination of the targets in relation to the original images, with the Gamma adaptive filter being the most efficient among the other tested filters. Texture image classifications were generally better than filtered image classifications, indicating that texture measures can be useful attributes to maximize discrimination of agricultural targets. The classes with the greatest discrimination potential in both F5A and S7D images, with accuracy above 50%, were: water, urban area, sugarcane-2, soy and exposed soil-1.