霉变小麦气相色谱–离子迁移谱的宽度学习检测模型

A broad learning detection model on gas chromatography-ion migration spectrum of mildew wheat

中文关键词  霉变小麦  气相色谱–离子迁移谱  指纹图谱  宽度学习  空间注意力

英文关键词  mildew wheat  gas chromatography-ion mobility spectrum  fingerprint spectrum  broad learning  spatial attention mechanism

中文摘要

      常用的小麦霉变检测方法存在检测程序复杂、环境适应性差等问题. 本文针对这一现状, 将具有高灵敏度的气相色谱–离子迁移谱(GC-IMS)应用于小麦的早期霉变检测, 对不同霉变程度的小麦样品进行GC-IMS测试并采用宽度学习模型(BLN)进行模式分类. 为了提高宽度学习模型的分类精度, 在模型中引入了空间注意力机制(SAM),通过使用节点的特征信息和结构信息计算注意力权重, 提取更重要的特征信息. 实验结果表明, 与现有的深度学习模型相比, 本文提出的模型在训练时间上大大减少, 在样本较少的情况下, 对霉变小麦早期识别的准确率(AUC)也得到了相应提高, 有效地解决了过拟合问题. 实验也证明了GC-IMS结合BLN-SAM模型的方法在小麦霉变早期检测中的有效性

英文摘要

      The commonly used methods for wheat mildew detection have some problems such as complicated detection procedures and poor environmental adaptability. Because of this situation, the high-sensitivity gas chromatography-ion migration spectrum (GC-IMS) was applied to detect early mildew of wheat, and the wheat samples with different mildew degrees were tested by using the GC-IMS and classified by using a broad learning model (BLN). To improve the classifi-cation accuracy of the broad learning model, a spatial attention mechanism (SAM) is introduced into the model, and the feature information and structure information of nodes are used to calculate attention weight and extract more important feature information. Experimental results show that compared with existing deep learning models, the training time of the proposed model is greatly reduced. In the case of a small number of samples, the early recognition accuracy (AUC) of mildew wheat was also improved, which effectively solved the problem of over-fitting. The experiment also proved the effectiveness of the GC-IMS combined with the BLN-SAM model in the early detection of wheat mildew.