Farmland detection in synthetic aperture radar images with texture signature

Abstract. The detection of farmland in synthetic aperture radar (SAR) images is useful to compute agriculture distribution in mountainous regions. The SAR technology is helpful to government agencies compiling much needed information for agricultural assessment of need-based data. We propose a texture signature to detect farmland in SAR. The texture signature is extracted from the texture pixels of the SAR image through the fuzzy c-means, where each texture pixel is a vector whose elements are the convolution value of the filters of the normalized Gaussian derivatives and SAR images at a spatial position. Then, we use the texture signatures to detect farmland in SAR images through the earth mover’s distance method. In the end, we propose a different approach to compute both the true positive rate and the false positive rate of receiver operating characteristic (ROC) curve. We use the area under the curve of ROC to achieve the best sample and the best threshold which realizes the best detection. The experiment results also show the best performance of the detection.