关键词: 水稻, 氮素营养诊断, 离散小波变换, 遗传算法, BP神经网络

Abstract: The genetic algorithm was used to optimize the BP neural network for rice nitrogen nutrition diagnosis, which provided theoretical guidance for the rational nitrogen application of rice. The rice variety used in field experiment was ‘Liangyoupei 9’, and four nitrogen levels (0, 210, 300, 390 kg·hm-2) were set. During the young panicle stage, the image of the third fully expanded leaf on the top of the rice was scanned, and the color and geometric features of the 19-dimensional rice images were obtained through image processing technology. Normalized processing, discrete wavelet transform and principal component analysis were used to preprocess the original data, and BP neural network optimized by genetic algorithm was used to diagnose nitrogen nutrition of rice. The rice nitrogen diagnosis model established by this method was better than the single BP neural network model and the traditional genetic algorithm optimized BP neural network model. The average recognition rates of the four nitrogen application levels obtained by the model tests were 100.000%, 99.000%, 97.000% and 100.000%, respectively, and the average total recognition rate of the test set samples reached 99.000%. The rice nitrogen nutrition diagnosis model established based on the genetic algorithm optimized BP neural network had strong learning ability and generalization ability, and could also well identify the lack of rice nitrogen nutrition, which indicated that the application of this method could be used to diagnose and identify nitrogen nutrition in rice very well.

Key words: rice, nitrogen nutrition diagnosis, discrete wavelet transform, genetic algorithm, BP neural network