ASSESMENT AND EVLUATION OF THE IMPACT OF USING POLSAR IMAGERIES WITH DIFERENT INCIDENT ANGLES IN FOREST CLASSIFICATION

Abstract. Forests are a dominant biome of the earth and have an important impact on its economic and environmental well-being. Forestry applications of radar remote sensing are addressed in the context of both forest management and ecosystem understanding, modelling and monitoring. Nowadays, radar remote sensing is being used for a lot of applications in various fields. Due to the applications of polarimetric radar in recent decades, many researchers have tended to this field. One of the main advantages of SAR images is that these images are independent over the time (day and night) and weather condition. The polarimetric SAR (POLSAR) images compared with other remote sensing images are more informative. Classification of radar images is a way by which we can separate different types of forest species. In addition to the main characteristics of the target, the backscatter from a SAR image is widely dependant on various radar system parameters. One of these system parameters is the incident angle of the radar system. In this paper, the impact of using PolSAR images with different incidence angles for the classification of forest areas is investigated. Two polSAR images with different incident angles taken by RADARSAT-2 in fine quad polarized mode (FQ4 and FQ18) have been used in this study. The study area is located in the Petawawa Research Forest (PRF) near Chalk River, Ontario, Canada The methodology of this paper contains three steps: (1) preprocessing, (2) wishart classification and (3) evaluating & analyzing the results. The preprocessing steps consist of the speckle noise filtering, covariance matrix extraction and georeferencing. In the second step, each incidence angle image was classified by using the supervised Wishart classification. The Wishart classification method has the capability of having multiple images at the same time. Thus, in the next experiment the classification was performed using both incidence angle images. Finally, the obtained results from each class and the classification results in each of three cases were evaluated and analyzed.

The results showed that Wishart classification provides an overall accuracy of % 67.17 using the lower incidence angle PolSAR image and % 65.38 for the higher incidence angle POLSAR image. Also, the overall accuracy of the simultaneous classification by using the extracted covariance matrix from both images is 72.63 %. Those results showed a better performance of the image with lower incident angle compared to that of an image with higher incident angle for forest classification. It is also shown that combining extracted covariance matrix from the FQ4 image with extracted covariance matrix from the FQ18 image can significantly improve the classification accuracy (overall accuracy).