Deep learning for multimedia data in IoT

Abstract

With the advent of Internet leading to proliferation of large amounts of multimedia data, the analytics of the aggregated multimedia data is proven to be one of the active areas of research and study. Multimedia data includes audio, video, images associated with applications like similarity searches, entity resolution, and classification. Visual data mining is now one of the active learning fields that include surveillance applications for object detection, fraud detection, crime detection, and other applications. Multimedia data mining includes many challenges like data volume, variety, and unstructured nature, nonstationary, and real time. It needs advanced processing capabilities to make decisions in near real time. The existing traditional database systems, data mining techniques cannot be used because of its limitations. Hence, to process such large amounts of data advanced techniques like machine learning, deep learning methods can be used. Multimedia data also includes sensor data that is widely generated. Most of the healthcare applications include sensors for detecting heart rate, blood pressure, and pulse rate. The advancement of the smartphones has resulted in fitness based applications based on the number of steps walked, calories count, kilometers ran, etc. All these types of data can be classified as Multimedia data for Internet of Things (IoT). There are many interfacing devices that are interconnected to each other with backbone as a computer network when sensor data is involved. The main aim of this chapter is to highlight the importance and convergence of deep learning techniques with IoT. Emphasis is laid on classification of IoT data using deep learning and the essential fine-tuning of parameters. A virtual sensor device implemented in python is used for simulation. An account of protocols used for communication of IoT devices is briefly discussed. A case study is provided regarding classification of Air Quality Dataset using deep learning techniques.