A Cloud Classification Model of Multi-spectrum Satellite Cloud Images Based on the Network Coupling SOFM with PNN
Huang Bing · Yingyong jichu yu gongcheng kexue xuebao · 2008
Based on the multi-spectrum samples of stationary meteorology satellite cloud images,the best distinguishing factors of cloud classification were distilled and a synthetical optimized cloud classifier combining the advantages of both self-organizing feature map (SOFM) and probabilistic neural network (PNN) was established by computing and analyzing the gray- gradient co-occurrence matrix and texture characters of satellite cloud image samples to overcome the shortcoming of single a neural network (ANN) classifier difficult to identify and classify complex cloud characters accurately and effectively.Firstly,the cloud samples were classified to identify and partition the analogical sample-sets by SOM without supervision,then the initial classification error was revised under the supervision and the initial classification results were optimized once more by using PNN.The experiments results showed that the synthetical SOM-PNN classifier can improve the distinguishing effect in that the total accuracy of cloud classification results could reach 92.4% and the coefficient of Kappa was 90.82, which excel other single-statistical classifier evidently.