Forest Classification in Mediterranean Areas Using a Quadratic Bayesian Classifier Method Applied to SAR Images

Abstract

<p class=MDPI16affiliation style=margin-left:0in;mso-add-space:auto; text-align:justify;text-indent:0in;line-height:normal;mso-list:l0 level1 lfo1> <span style=font-size:11.0pt;font-family: new= roman,serif;color:black;mso-themecolor:text1=>1. Abstract <p class=MDPI17abstractCxSpLast style=margin:0in;margin-bottom:.0001pt; mso-add-space:auto;line-height:normal><span style=mso-bidi-font-size:10.0pt; font-family: new= roman,serif;color:black;mso-themecolor:text1=>In this paper, multifrequency SAR images from ALOS/PALSAR, ENVISAT/ASAR, and Cosmo-SkyMed sensors have been studied for highlighting forest features in a test area in Central Italy (San Rossore), where detailed in-situ measurements were available. A preliminary discrimination of the main land cover classes and forest types was carried out by exploiting the synergy among various frequencies (L, C and X bands). After a pre-processing of the available SAR images by applying a multilook approach, SAR data were used for discriminating forest from non-forest land covers and separating broadleaved from coniferous forest types by using RGB compositions of multi-temporal and multi-frequency images. The mean backscattering coefficient ( <span style=mso-bidi-font-size: 10.0pt;font-family:Symbol;mso-ascii-font-family: new= roman;mso-hansi-font-family:= times= roman;color:black;mso-themecolor:text1;mso-char-type:symbol;= mso-symbol-font-family:symbol=>s °) was computed for each sensor and available polarization from the pixels associated to coniferous and broadleaf obtained from the reference classification map. The classification has then been performed by applying to the SAR images, in different configurations of polarizations and frequencies, a new method based on a quadratic Bayesian classifier, which is able to overcome the limits of ground-truth classes that contain not homogenous targets (i.e. non-forest class). The obtained results indicated that the different surface types were best identified by the joint use of X and L bands (the correct classification show 80.13%, 83.03% and 75.07% for coniferous, broadleaf forests and non-forest respectively). The best overall accuracy is also obtained by considering the joint use of L and X bands (80.06%). <p class=MDPI18keywordsCxSpLast style=margin-top:0in;margin-right:0in; margin-bottom:0in;margin-left:.25in;margin-bottom:.0001pt;mso-add-space:auto; text-indent:-.25in;line-height:normal;mso-list:l0 level1 lfo1> <span style=font-size:11.0pt;font-family: new= roman,serif;color:black;mso-themecolor:text1=>2. Keywords : <span style=mso-bidi-font-size: 10.0pt;font-family: new= roman,serif;color:black;mso-themecolor:text1=>Bayesian Classifier; Forest Features; Forest/Non-Forest Areas; Land Classification; SAR