Evaluation of oak wilt index based on genetic programming

We proposed a normalized oak wilt index (NWI) to extract oak wilt area from remotely sensed hyperspectral image of forest in our previous work. The NWI, which is designed based on factitious characterization of spectral profiles of oak wilt, realized satisfactory extraction performance. In this paper, we propose a genetic-programming-based search method for physically interpretable index. The search procedure consists of two stages, i.e. extraction for significant binary operations and tree construction, in expectation of dealing with more subtle problem, e.g. estimation of quantities of ingredients of vegetation. The selected binary operations are consistent with plant physiology. The extraction performance of proposed method based on fewer binary operations stands comparison with NWI´s performance.