Extending self-organizing maps for supervised classification of remotely sensed data

An extended self-organizing map for supervised classification is proposed in this paper. Unlike other traditional SOMs, the model has an input layer, a Kohonen layer, and an output layer. The number of neurons in the input layer depends on the dimensionality of input patterns. The number of neurons in the output layer equals the number of the desired classes. The number of neurons in the Kohonen layer may be a few to several thousands, which depends on the complexity of classification problems and the classification precision. Each training sample is expressed by a pair of vectors: an input vector and a class codebook vector. When a training sample is input into the model, Kohonens competitive learning rule is applied to selecting the winning neuron from the Kohonen layer and the weight coefficients connecting all the neurons in the input layer with both the winning neuron and its neighbors in the Kohonen layer are modified to be closer to the input vector, and those connecting all the neurons around the winning neuron within a certain diameter in the Kohonen layer with all the neurons in the output layer are adjusted to be closer to the class codebook vector. If the number of training samples is sufficiently large and the learning epochs iterate enough times, the model will be able to serve as a supervised classifier. The model has been tentatively applied to the supervised classification of multispectral remotely sensed data. The author compared the performances of the extended SOM and BPN in remotely sensed data classification. The investigation manifests that the extended SOM is feasible for supervised classification.     

Review of Remotely Sensed Imagery Classification Patterns Based on Object-oriented Image Analysis

1 Introduction With the rapid development of remotely sensed (RS) information collection, transfer and storage in the last two decades, the limitation of RS application is becom- ing weaker because of availability of multiple RS data sources of increasingly finer spatial, temporal, spectral and radiant dimensions. In the high spatial-resolution RS imagery, characteristics of land-cover are fairly clear such as spatial shape, structure, texture, etc., so the mixture of different land covers …     

多源遥感影像数据融合方法在地学中的应用

应用遥感影像数据融合理论，研究了光学遥感、热红外遥感、微波遥感卫星在地质学中的应用，论述了不同类型的遥感数据及其他地学数据在不同层次上的同一传感器多波段数据融合、不同SAR图像数据融合、HIS变换、遥感影像与地球物理、地质和航磁等数据的融合。结果表明，数据融合技术在突出地质特征信息方面具有能突出线性构造、断裂构造、地形地貌的优势。     

Integrating TM and ancillary geographical data with classification trees for land cover classification of marsh area

The main objective of this research is to determine the capacity of land cover classification combining spectral and textural features of Landsat TM imagery with ancillary geographical data in wetlands of the Sanjiang Plain, Heilongjiang Province, China. Semi-variograms and Z-test value were calculated to assess the separability of grey-level co-occurrence texture measures to maximize the difference between land cover types. The degree of spatial autocorrelation showed that window sizes of 3×3 pixels and 11×11 pixels were most appropriate for Landsat TM image texture calculations. The texture analysis showed that co-occurrence entropy, dissimilarity, and variance texture measures, derived from the Landsat TM spectrum bands and vegetation indices provided the most significant statistical differentiation between land cover types. Subsequently, a Classification and Regression Tree (CART) algorithm was applied to three different combinations of predictors: 1) TM imagery alone (TM-only); 2) TM imagery plus image texture (TM+TXT model); and 3) all predictors including TM imagery, image texture and additional ancillary GIS information (TM+TXT+GIS model). Compared with traditional Maximum Likelihood Classification (MLC) supervised classification, three classification trees predictive models reduced the overall error rate significantly. Image texture measures and ancillary geographical variables depressed the speckle noise effectively and reduced classification error rate of marsh obviously. For classification trees model making use of all available predictors, omission error rate was 12.90% and commission error rate was 10.99% for marsh. The developed method is portable, relatively easy to implement and should be applicable in other settings and over larger extents.