Cloud Classification Based on Self-Organizing Feature Map and Probabilistic Neural Network

For overcoming the shortcoming of single ANN classifier being difficult used to identify and classify complex clouds, based on the multi-spectrum samples of stationary meteorology satellite cloud pictures, by computing and analysing the gray-grads and texture characters of satellite cloud picture samples, and picking-up some suitable cloud classification factors, a synthetic optimization SOM-PNN cloud classifier was designed and established. Firstly, the clouds samples were classified by SOM in the form of unsupervised, by identifying and partitioning the analogical sample-sets and getting rid of the invalid data, then the initial classification error were revised and the initial classification results were more optimized by using PNN in the form of supervised. The experiments results of cloud classification showed that the accuracy of the synthetic SOM-PNN classifier can achieved 90%, which is evidently excel to other single-statistical cloud classifier