Neural network clustering method for microwave radiometer remote-sensing atmospheric parameter

The invention relates to a neural network clustering method for a microwave radiometer remote-sensing atmospheric parameter. The method is characterized by comprising the following steps: firstly, using historic sounding data to quantitatively calculating correlations among different stratification of atmosphere to select the stratification with a large correlation coefficient, substituting the stratification into an artificial neural network through clustering for training, and inverting to output atmospheric parameters of different clusters respectively. According to method, the weight and deviation of the network are constantly adjusted by an artificial neural network algorithm to reduce the deviation between an output vapor density vector obtained through input vector calculation and actual training target output vapor density vector. After the training is completed, the weight and deviation of the network are determined, therefore, during practical observation by a microwave radiometer, all that needed is to call a weight matrix and deviation, the operating speed is higher and stable, and higher inversion accuracy is achieved.