Anomaly Detection in Particulate Matter Sensor Using Hypothesis Pruning Generative Adversarial Network

Abstract

World Health Organization (WHO) provides the guideline for managing the Particulate Matter (PM) level because when the PM level is higher, it threats the human health. For managing PM level, the procedure for measuring PM value is needed firstly. We use Tapered Element Oscillating Microbalance (TEOM)-based PM measuring sensors because it shows higher cost-effectiveness than Beta Attenuation Monitor (BAM)-based sensor. However, TEOM-based sensor has higher probability of malfunctioning than BAM-based sensor. In this paper, we call the overall malfunction as an anomaly, and we aim to detect anomalies for the maintenance of PM measuring sensors. We propose a novel architecture for solving the above aim that named as Hypothesis Pruning Generative Adversarial Network (HP-GAN). We experimentally compare the several anomaly detection architectures to certify ours performing better.

Multimodal layer surveillance map based on anomaly detection using multi‐agents for smart city security

“…Using the deep learning algorithm, it is possible to determine whether the current situation is normal [23]. The proposed surveillance map comprises an observation map and a probability map, and unusual situations can be estimated [24].…”

Non-Compression Auto-Encoder for Detecting Road Surface Abnormality via Vehicle Driving Noise

“…One of them is RNN based model and the other one is based on convolutional neural networks (CNN). For example, FARED and HP-GAN are developed based on RNN and CNN respectively for anomaly detection of time series data [4,5].…”

“…For easing the above limitation, CNN can be considered for constructing neural network architecture as shown in HP-GAN case [5]. The convolutional filter sliding on the input data and aggregating the spatial information for generating result.…”

Efficient Non‐Compression Auto‐Encoder for Driving Noise‐Based Road Surface Anomaly Detection

“…Recently, many studies use generative neural networks as anomaly detection models [12–26]. The general premise of those anomaly detection models is that model, learned only normal state data, cannot generate abnormal state data well.…”

“…One of the reasons is using variational bound and the other is parameter sharing between encoder and decoder [20–24]. The above limitation is not only solved by using feature matching and multiple hypotheses but also improved anomaly detection performance [25,26]. However, the limitation of those is the model becomes more complex.…”