Study on probabilistic neural network for extracting remote sensing information of rice planting area
概率神经网络的水稻种植面积遥感信息提取研究

In order to improve the extraction precision of rice planting area,multitemporary remote sensing images chosen based on the growth stages of rice were performed atmospheric correction and geometric rectification.The fusion algorithms which are used to select the optimal bands combination include single band statistic,principal component transformation and ratio transformation.The basic algorithm and theory of the PNN(probabilistic neural network) were analyzed,and it was applied to classify the image of the optimal bands combination.The classified result was compared with those of BP(back propagation) neural network and minimum-distance method.Results show that the classification precision of PNN is higher than that of minimum-distance by 6 percentage points,and BP by 13 percentage points.As for the precision of rice planting area extraction,the PNN's extraction precision is higher than that of minimum-distance by 15 percent points.Therefore,PNN is an effective method for classification of remote sensing images,and it plays a unique role in extracting the crops planting area.