Based on the visible-near infrared reflectance spectra of broad bean leaves, by combining the derivative spectra, we analyzed the spectral characteristics of experiment samples with three levels of pests: healthy leaf, leaf with a small amount of pests and leaf with many pests, and selected the optimum waveband for pest detection. The Hadoop, Spark and VMWare virtual machines were used to build the cloud computing platform, and the MLlib machine learning library was used to realize the classification models of artificial neural network (ANN) and support vector machine (SVM). We also conducted classification modeling and prediction of the full waveband and optimum waveband spectra of broad bean leaves with three levels of pests. The experiment results showed that the ANN pest spectrum classification model had higher accuracy than the SVM pest spectrum classification model, and the ANN model also had higher operating efficiency on the cloud platform.In the meantime, the prediction accuracy for full-waveband spectrum was higher than that for optimum waveband. By expanding the spectrum datasets, the computational efficiency of clouding computing technology in spectrum data mining can be significantly improved. The classification detection based on cloud computing can provide new technique and method for the spectral recognition of crop biotic stress.