Automatic determination of corrosion severity is an important task that has not received adequate study due to the non-availability of datasets. This study explores corrosion severity detection by leveraging representation learning for the development of lightweight, shallow machine learning models. A variational autoencoder was used for feature extraction. Four classifiers, specifically RF, SVC, k-NN, and XGBoost, were trained using the encoded representation without any additional processing. Voting classifiers were also constructed using the trained models. Except for the slight drop in precision (6.06%) of the soft voting classifier, augmentation produces a general positive influence on the precision, recall, and F1-score of the models understudied; it improved the precision, recall, and F1-metrics of K-NN and hard-voting classifiers most remarkably. For K-NN, the improvement reached 29.17%, 20.93%, and 30.77% for precision, recall, and F1-metrics, respectively, and 30.36%, 26.42%, and 35.29%, respectively, for the hard voting classifier. Whereas our experiments could not produce stateof-the art results, it provides adequate motivation to further study VAE as a data preprocessing unit for the development of simple, efficient, lightweight models that can be deployed on resource constrained devices, which will in turn advance the development and deployment of corrosion monitoring systems on low-cost devices.