基于激光诱导击穿光谱积分信号的水稻种子鉴别研究

基于LIBS技术结合BP神经网络技术,对6类水稻种子进行类型鉴别研究。对水稻种子的LIBS全谱和分段光谱进行积分,再输入BP神经网络,得到:全谱积分前的识别率为81.02%,积分后识别率为93.40%,识别率提高12.38%;分段光谱积分后识别效果较好的是中心波长为405 nm、570 nm、810 nm的光谱,将这三段光谱进行组合,组合光谱的识别率均超过94%,其中405-570-810 nm组合识别率最高达到97.67%;从该组合光谱中提取特征光谱输入神经网络,识别率为97.35%,识别时间为原组合光谱的53%。结果表明:LIBS积分信号下的组合光谱法识别迅速可靠,适用于水稻种子的快速批量检测。

Identification of rice seeds based on laser-induced breakdown spectral integral signal

The Laser-Induced Breakdown Spectral integral signal combined with the Back Propagation neural network technology was applied to identify rice seeds types.Integrate the laser-induced breakdown full-spectrum and segmented spectral lines of rice seed samples,and input the spectral data before and after integration into the BP neural network,the result is as follows:The identification rate of rice seeds before full-spectrum integration is 82.02%,the identification rate after integration is 93.40%,the recognition rate has increased by 12.38%,and the recognition time is longer.The integrated segmented spectrum has a better identification effect,and the centre wavelength is405 nm,570 nm,and 810 nm.Combining these three spectrums,the recognition rate after the combined spectrum integration exceeds 94%,of which the highest recognition rate of the 405-570-810 combination reaches 97.67%and the recognition time is 64.17%.The feature spectrum was extracted from the 405-570-810 nm combined spectrum and input to the neural network.The recognition rate is 97.35%,and the recognition time was significantly reduced,which is 53% of the original combined spectrum.The results show that the combined spectroscopy method under the LIBS integrated signal is fast and reliable,and is suitable for rapid batch detection of rice seeds.