Rapid THz Identification of Coffee Bean Origin with Ensemble Learning

In this study, a classification model for THz spectral data of coffee is constructed using an integrated learning approach, an AELM optimization model is proposed, the ELM is improved using the AO population optimization algorithm, the connection weights of the input and implicit layers of the ELM and the threshold of the implicit layer are searched for, the AELM is used as a weak classifier of FSAMME for integrated learning, the weights of the FSAMME algorithm are improved The update method is used to increase the weight of misclassified sample data and reduce the weight of weak classifiers with high classification error rate in the final classifier by dynamically weighting them during the iteration process according to the classification accuracy, and finally normalize all weak classifier weights to integrate the strong classifier AE-dynamic FS integrated learning model. The accuracy of AO-ELM-dynamic FSAMME model on the test set sample data set of five coffee origins is 99%, the classification accuracy of coffee samples from China, Brazil, Colombia, Ethiopia and Honduras is 100%, 100%, 100%, 94.4% and 100% respectively, and the number of samples misclassified is 1 sample from Ethiopia,realizing excellent classification performance.