Uncertain Probabilistic Machine Learning Algorithm for Scene Classification of Humanoid Robot

Scene classification has been standing as an important task for many practical applications. In particular, noisy existence, different view angle, variations in lighting conditions, and dynamic backgrounds lead to obstacles in efficient learning and robust classification. Stochastic and uncertain information is always ignored which is existing all time in the practical systems. In this paper, we propose a novel Uncertain Probabilistic Machine Learning Algorithm (UPMLA) for scene classification in humanoid robot system to address these difficulties. Firstly, uncertain probabilistic theorem is integrated with Support Vector Machines to construct Uncertain Probabilistic Fuzzy Support Vector Machines (UPFSVMs) model. Secondly, in order to obtain the solutions of the UPFSVMs, we can transform the probabilistic fuzzy convex optimization problem to the deterministic convex optimization problems. As a result, the solutions of the UPFSVMs can be represented by a set of sub-hyperplanes with the Probabilistic Fuzzy Degree (PDF). Therefore, the optimization problem of the UPFSVMs essentially revolves around designing the PFD. Finally, the design of PFD is predicted on statistical confidence-entropy, which enhance the robustness of the classification process. Compared to other popular scene classification algorithms, experimental results demonstrate the competitiveness of the proposed UPMLA algorithm. Specifically, the UPMLA achieves superior comprehensive metrics within a similar computing budget on the newly built ‘Alice’ datasets and OT data set.