Pedestrian Classification Using Self-Training Algorithm

Deep semi‐supervised classification based in deep clustering and cross‐entropy

“…In a number of studies, semi‐supervised approaches, including self‐training, 9 co‐training, 12 generative models, 13 graph‐based methods, 14 and semi‐supervised support vector machines 15 have been employed. One of the simplest and most efficient semi‐supervised models is the self‐training approach, 9 which is widely used in several domains, such as object detection, facial recognition, EGG signal classification, and time series problems 16–19 …”

“…One of the simplest and most efficient semi-supervised models is the selftraining approach, 9 which is widely used in several domains, such as object detection, facial recognition, EGG signal classification, and time series problems. [16][17][18][19] The self-training process is simple and straightforward. First, the labeled data are used to train a classifier, which then assigns the predicted class labels to the unlabeled data.…”

“…summarizes the features of the databases used. All databases are normalized using the MinMax method defined in Equation(17), where v, v′, min A , and max A are the original, normalized, minimum, and maximum values of the analyzed feature, respectively.In this study, two performance measures that are commonly used because of their simplicity and successful application to multi-class classification problems are employed. As classification Details of data sets used in experiments…”