基于深度学习的烟丝杂物多光谱检测技术

Machine learning-based multispectral detection of tobacco impurities

针对烟丝杂物剔除问题,提出了基于深度学习的烟丝杂物多光谱检测方法。首先设计了总体检测方案和检测系统,搭建了多光谱实验验证系统。利用高光谱成像技术和深度学习方法对烟丝中的非烟物质进行分类识别,针对杂物和烟丝对光的反射特性不同,利用光谱相机拍摄不同波段的图片,分别比较机器学习算法和深度学习算法进行的训练与分类,通过GoogLeNet深度学习算法识别烟丝中的杂物。结果表明:检测识别与剔除正确率为99.5%,证实了检测技术的高效性和准确性。

A multi-spectral detection method for tobacco impurities based on deep learning was proposed to solve the problem of removing tobacco impurities.Firstly,the overall detection scheme and detection system are designed,and the multispectral experimental verification system is built.Hyperspectral imaging technology and deep learning method were used to classify and identify nontobacco substances in tobacco leaves.In view of the different reflection characteristics of impurities and tobacco,spectral cameras were used to take pictures of different bands,and machine learning and deep learning algorithms were compared for training and classification.The GoogLeNet deep learning algorithm was used to identify the impurities in tobacco,and the detection accuracy reached 99.5%,proving the detection technology's high efficiency and accuracy.