Estimation of Aboveground Biomass from Satellite Data Using Quaternion-Based Texture Analysis of Multi Chromatic Images

In recent years, first approaches using quaternion numbers to handle and model multi chromatic images in a holistic manner were introduced. By defining quaternion Fourier transform, multidimensional data such as color images can be efficiently and easily process. On the other hand, multi chromatic satellite data appear as a primary source for measuring past trends and monitoring changes in forest carbon stocks. Thus, the processing of these data represents a fundamental challenge. In this work, inspired by the quaternion Fourier transforms, we propose a texture-color descriptor to extract relevant information from multi chromatic satellite images. We also propose a quaternion-based texture model, named FOTO++, to address the aboveground biomass estimation issue. Our proposed model begins by removing noises in the multi chromatic data while preserving the edges of canopies. After that, color texture indices are extracted using discrete form of Quaternion Fourier Transform and finally support vector regression method is used to derive biomass estimation from texture indices. Our texture features are modeled by a vector composed by the radial spectrum coming from the amplitude of quaternion Fourier Transform. We conduct several experiments in order the study the sensitivity of our model to acquisition parameters. We also assess its performances both on synthetic images and on real multi chromatic images of Cameroonian forest. The results provided support that our model is more robust to acquisition parameters than the classical Fourier Texture Ordination model and it is more accurate for aboveground biomass estimates. We stress that similar methodology could be used with quaternion wavelets. These results highlight the potential of quaternion-based approach to study multi chromatic images.