A Novel Multiple Sensors Detection Framework via Feature Combination Selection and Weighted Modified Adaptive Graph Learning

In this paper, a multi-sensor detection framework for bacterial infection types recognition is developed, including a sensor array with multiple types of gas sensors, a compatible feature combination selection framework, and an array optimization algorithm. To capture more information from the sensor response curves, such features include but are not limited to maximum response value, maximum first-order derivative, and the area under the response curve are extracted from each sensor response curve. Additionally, a classifier-based feature combination selection method is employed to form a feature combination that contributes significantly to the classification performance. Since the broad-spectrum response and the cross-sensitivity characteristics of the sensors and the possible overlapping selectivity of the sensor array can bring about redundant and interference information, a novel array optimization algorithm that can adaptively learn the structure of the data is developed, which leverages the importance of the features in the classification task as well as the intrinsic structure among the samples in the projection space. The classification results of different classifiers verify the effectiveness of the designed system. Comparative analysis with existing algorithms verifies the effectiveness of the developed array optimization framework.