Semi-supervised Anomaly Detection based on Improved Adversarial Autoencoder and Ensemble Learning
Wenfen Liu, Haoyang Jia, Nan Wang, Yuehua Huang · 2021 4th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE) · 2021
We propose a novel anomaly detection method based on improved adversarial autoencoder and ensemble learning. Our method improves on the traditional adversarial autoencoder, and adds an encoder to extract additional feature vectors. The first set of feature vectors is extracted from the input image by the first encoder, and the output image is reconstructed by the decoder. Additional encoders use this reconstructed image to extract a second set of feature vectors. Two sets of feature vectors are weighted and fused to form a new set of feature vectors. Taking this new set of feature vectors as training data, AdaBoost algorithm is used to train and integrate several single-layer decision trees, and finally a strong classifier is obtained for anomaly detection. The proposed method has been verified and evaluated on several published image datasets, and has achieved excellent performance compared with the previous methods.