The weedy rice and rice in growth period of 10d were investigated. The hyper spectral image data were captured from weedy rice and rice leaves. After image data were filtered, the feature images at wavelength of 1448.89nm and 1469.89nm were optimized by principal component analysis method. For each feature image, shape feature, texture feature and color feature were extracted, and 18 feature variables in all were attained. Neural network method was used to build the discriminate model. The discriminating rates for weedy rice and rice were both 100% in training set. The discriminating rate for weedy rice was 92.86% and the discriminating rate for rice was 96.88% in prediction set. Experimental results showed that the hyper-spectral imaging technology could be used to identify weedy rice and rice at seeding stage.