Laboratory Safety Anomaly Detection Based on GAN

For the sake of better monitor the laboratory security problem, this thesis introduces the outlier detection technology into the laboratory security detection, and proposes an unsupervised anomaly detection algorithm on account of generative adversarial network. The algorithm is trained with positive samples to detect laboratory safety anomalies, and solves the problems of difficult sample acquisition and high labeling cost. To settle the training question of generating adversarial networks, the optimizer is optimized and the learning rate parameters are adjusted adaptively. In order to make the network to be more cautious about the main areas in the laboratory, the channel space hybrid attention module CBAM is introduced to help the model be more cautious about abnormal areas in the detection process. Experimental results on self-made data sets diaplay that proposed method can enhance the anomaly test ability of the detection and improve the training stability.