干旱区生态环境知识资源中心

Macrostructure/Wavelet-texture-label per pixel: In this paper, we concentrate on neural network classifiers on sub;regions of the image and we show how texture information obtained with a wavelet transform can be integrated to improve such a single label classifier. We apply a local spatial frequency analysis, a wavelet transform, to account for statistical texture information in Landsat/TM imagery. Statistical texture is extracted with a continuous edge-texture composite wavelet transform. We show how this approach relates to texture information computed fi om a co-occurrence matrix. The network is then trained with both the texture information and the additional pixel labels provided by the ground truth data. Theory and regional results are described in this paper.

Microstructure/ICA-percentage of each pixel: Based on the experience of the deforestation surveillance (cf. this Proceedings Vol 3723, plenary paper on Progresses of Unsupervised Neural Networks, Part II Real World Applications), we need to know at all the forest boundary pixels the percentage of tree content of each pixel, i.e. multiple labels with percentages. Thus, in this paper, we generalize the single label classifier to multiple labels. ICA-ANN discovers not only the not yet cool down sensor noise of TM-1 to TM-4 flying over Maryland Charles County near Washington DC, and the realistic farm sub-structure, kibbutz, in the desert settlement of of Israel.