基于近红外光谱技术及ELM对小麦中不同生长阶段米象的分类识别

Classification and recognition of Sitophilus oryzae in different growth stages of wheat based on near-infrared spectroscopy and ELM

对粮食中隐蔽性害虫的早期诊断和检测,不仅可以减少因害虫取食造成的粮食产后损失,还可以减少化学药剂的使用,对于保证粮食品质和减少环境污染具有重要的意义。基于近红外光谱技术与极限学习机(ELM)构建小麦中不同生长阶段米象的分类识别模型,采集未感染小麦和感染米象小麦的近红外光谱数据,选择SNV+De-trending的组合对原始光谱数据进行预处理,使用主成分分析(PCA)方法对光谱数据进行降维特征提取,利用ELM和支持向量机(SVM)建立分类识别模型。结果表明:ELM模型训练时间仅需0.062 5 s,总体分类准确率为90%,0、6、24和27 d的识别率为100%,10~20 d的幼虫期识别率偏低,20 d时识别率最低,为65%;SVM模型运行时间为3.38 s,分类准确率为85.42%,ELM模型较SVM模型的运行时间和分类准确率都有所提高。因此,ELM分类识别模型能够快速准确地判断小麦有无米象,以及分类识别小麦中不同生长发育阶段的米象。

Early diagnosis and detection of hidden pests in grain could not only reduce the post-production losses of grain caused by pest feeding, but also reduce the use of chemicals, which are important for maintaining grain quality and reducing environmental pollution. In this paper, a classification and identification model of Sitophilus oryzae(S. oryzae) in wheat at different growth stages was constructed based on near-infrared spectroscopy and extreme learning machine(ELM), and the near-infrared spectral data of uninfected wheat and infected S. oryzae wheat were collected. The images of S. oryzae in different growth and development stages were obtained by X-ray imaging technology, and the development period of S. oryzae was obtained through the images(egg stage at 0-9 d, larval stage at 10-20 d, pupa stage at 21-26 d, and adult stage at 27-30 d), the wheat with full grains was selected and collected by near infrared spectroscopy to obtain the spectral data of uninfected samples, and then wheat was infested with S. oryzae adults. After 48 hours, the S. oryzae adults were taken out, and the samples were collected by near-infrared spectrum on 6, 10, 14, 17, 20, 24, and 27 day of the experiment to obtain uninfected wheat and near-infrared spectral data of wheat infected with S. oryzae wheat. When modeling using the original spectral data, the classification accuracy of the ELM model was 78.75%. After preprocessing, the classification accuracy of the ELM model reached 85%, and then the principal component analysis(PCA) method was used to perform dimension reduction feature extraction on the spectral data. When the target dimension was 120 dimensions, the accuracy of the ELM classification and recognition model was 90%, the classification recognition rate increased by 12.5%. The experimental results showed that the appropriate preprocessing method and PCA dimensionality reduction feature extraction could effectively improve the classification accuracy of ELM model, and the training time was only 0.062 5 s, the overall classification accuracy reached 90%, the recognition rates were 100% on 0, 6, 24, and 27 days, and lower on 10-20 days of larval stage, and the recognition rate was the lowest at 20 days, which was 65%. Compared with the performance of ELM and SVM in this experiment, the training time of SVM model was 3.38 s, the overall classification accuracy reached 85.42%, on 0, 10, 17, and 24 days, the recognition rate was 90%, at 6 days, the recognition rate was 80%, at 14 days, the recognition rate was 55%, and at 20 and 27 days, the recognition rate was 85%. The results showed that the classification effect of ELM model was better than that of SVM model. Therefore, ELM classification and recognition model could quickly and accurately determine insect-free and insect-containing wheat and classify the S. oryzae at different growth and development stages. The classification and identification of S. oryzae has potential practical value for early detection of hidden pests in grain. The data of this study comes from laboratory conditions. In the future, more data can be collected from actual production to strengthen the classification model and increase the accuracy of the model. The optimized ELM was used to further improve the classification and recognition efficiency and accuracy of the model. On the basis of this paper, a classification and identification model of a variety of hidden pests was established to provide a reference for the intelligent detection of pests in the construction of intelligent grain depots.