M.N. Favorskaya, N.Y. Petukhov, I.M. Danilin, and A.I. Danilin (Russia)
Remote sensing, texture, neural networks, fractal, statistic descriptors
Recognition of forest and its state on airphotos is one of important problems of natural resources monitoring. Automatic interpretation of forest textures photos is also a complex task which isn’t finally solved. In this paper we propose new method of forest textures recognition based on two-level procedure: (1) the pre-segmentation of airphoto based on image pyramid and definition of statistical similarity regions, and (2) the texture recognition using neural network of direct propagation with input complex fractal and statistical descriptors and the post-segmentation of airphoto. Application of additional methods of laser scanning permits to recognize trees using not only upper crowns but also their lateral surfaces. Thereby we can estimate morphological descriptors of leaves mass of trees analyzing the set of airphotos.
Important Links:
Go Back
