论文部分内容阅读
为了实现大豆品种的快速无损鉴别,对16份大豆品种的近红外透射光谱(NITS)进行分析。首先通过平滑和马氏距离的光谱预处理方法消除噪声和去除奇异光谱。然后分别用主成分分析(PCA)和离散多带小波变换(DWT)提取光谱特征,作为BP神经网络的输入,构建PCA-BP和DWT-BP大豆品种识别模型。结果表明:PCA-BP模型的识别准确率为98.125%,平均识别时间为9.3 ms;DWT-BP模型的识别准确率为95.93%,平均识别时间为6.4 ms。研究结果为大豆品种的快速无损鉴别提供了理论依据和实用方法。
In order to achieve rapid and non-destructive identification of soybean varieties, NITS of 16 soybean cultivars were analyzed. Firstly, the noise and the singularity spectrum are eliminated through the spectral preprocessing method of smoothing and Mahalanobis distance. Then PCA and DWT were used respectively to extract spectral features, which were used as input of BP neural network to construct PCA-BP and DWT-BP soybean variety identification models. The results showed that the recognition accuracy of PCA-BP model was 98.125% and the average recognition time was 9.3 ms. The recognition accuracy of DWT-BP model was 95.93% and the average recognition time was 6.4 ms. The results provided a theoretical basis and practical methods for rapid and nondestructive identification of soybean varieties.