مقاله انگلیسی رایگان در مورد یادگیری عمیق برای تشخیص فعالیت بر اساس حسگر – الزویر ۲۰۱۸

Human activity recognition (HAR) plays an important role in people’s daily life for its competence in learning profound high-level knowledge about human activity from raw sensor inputs. Successful HAR applications include home behavior analysis (Vepakomma et al., 2015), video surveillance (Qin et al., 2016), gait analysis (Hammerla et al., 2016), and gesture recognition (Kim and Toomajian, 2016). There are mainly two types of HAR: video-based HAR and sensor-based HAR (Cook et al., 2013). Video-based HAR analyzes videos or images containing human motions from the camera, while sensor-based HAR focuses on the motion data from smart sensors such as an accelerometer, gyroscope, Bluetooth, sound sensors and so on. Due to the thriving development of sensor technology and pervasive computing, sensor-based HAR is becoming more popular and widely used with privacy well protected. Therefore, in this paper, our main focus is on sensor-based HAR. HAR can be treated as a typical pattern recognition (PR) problem. Conventional PR approaches have made tremendous progress on HAR by adopting machine learning algorithms such as decision tree, support vector machine, naive Bayes, and hidden Markov models (Lara and Labrador, 2013). It is no wonder that in some controlled environments where there are only a few labeled data or certain domain knowledge is required (e.g. some disease issues), conventional PR methods are fully capable of achieving satisfying results. However, in most daily HAR tasks, those methods may heavily rely on heuristic handcrafted feature extraction, which is usually limited by human domain knowledge (Bengio, 2013). Furthermore, only shallow features can be learned by those approaches (Yang et al., 2015), leading to undermined performance for unsupervised and incremental tasks. Due to those limitations, the performances of conventional PR methods are restricted regarding classification accuracy and model generalization. Recent years have witnessed the fast development and advancement of deep learning, which achieves unparalleled performance in many areas such as visual object recognition, natural language processing, and logic reasoning (LeCun et al., 2015). Different from traditional PR methods, deep learning can largely relieve the effort on designing features and can learn much more high-level and meaningful features by training an end-to-end neural network.