Fundamentals Of Neural Networks Architectures Algorithms And Applications

McCabe's cyclomatic complexity metric (1976) is widely cited as a useful predictor of various software attributes such as reliability and development effort This critique demonstrates that it is based upon poor theoretical foundations and an inadequate model of software development The argument that the metric provides the developer with a useful engineering approximation is not borne out by the empirical evidence Furthermore, it would appear that for a large class of software it is no more than a proxy for, and in many cases is outperformed by, lines of code<
 >

A critique of cyclomatic complexity as a software metric

Multi-sensor information fusion based on machine learning for real applications in human activity recognition: State-of-the-art and research challenges

Human Activity Recognition (HAR) employing inertial motion data has gained considerable momentum in recent years, both in research and industrial applications. From the abstract perspective, this has been driven by an acceleration in the building of intelligent and smart environments and systems that cover all aspects of human life including healthcare, sports, manufacturing, commerce, etc. Such environments and systems necessitate and subsume activity recognition, aimed at recognizing the actions, characteristics, and goals of one or more individuals from a temporal series of observations streamed from one or more sensors. Due to the reliance of conventional Machine Learning (ML) techniques on handcrafted features in the extraction process, current research suggests that deep-learning approaches are more applicable to automated feature extraction from raw sensor data. In this work, the generic HAR framework for smartphone sensor data is proposed, based on Long Short-Term Memory (LSTM) networks for time-series domains. Four baseline LSTM networks are comparatively studied to analyze the impact of using different kinds of smartphone sensor data. In addition, a hybrid LSTM network called 4-layer CNN-LSTM is proposed to improve recognition performance. The HAR method is evaluated on a public smartphone-based dataset of UCI-HAR through various combinations of sample generation processes (OW and NOW) and validation protocols (10-fold and LOSO cross validation). Moreover, Bayesian optimization techniques are used in this study since they are advantageous for tuning the hyperparameters of each LSTM network. The experimental results indicate that the proposed 4-layer CNN-LSTM network performs well in activity recognition, enhancing the average accuracy by up to 2.24% compared to prior state-of-the-art approaches.

https://www.mdpi.com/1424-8220/21/5/1636/pdf

LSTM Networks Using Smartphone Data for Sensor-Based Human Activity Recognition in Smart Homes

Nowadays, one of the most important objectives in health-related research is the improvement of the living condition and well-being of people. Smart home systems can provide health protection for residents based on the results of daily activity recognition. Recent advances and developments in sensor technology have increased the need for sensor-compatible goods and services in smart homes. Consequently, the ever-increasing volume of data requires the field of deep learning (DL) for auto-matic human motion recognition. Recent research has modeled spatiotemporal sequences gathered by smart home sensors using long short-term memory networks. In this work, ResNeXt-based models that learn to classify human activities in smart homes were proposed to improve recognition performance. Experiments conducted on Center for Advanced Studies in Adaptive Systems (CASAS) data, a publicly available benchmark dataset, shows that the proposed ResNeXt-based techniques are significantly superior to the existing DL methods and provide better results compared to the existing literature. The ResNeXt model achieved the averaged accuracy over the benchmark method to 84.81%, 93.57%, and 90.38% for the CASAS_Cairo, CASAS_Milan and CASAS_Kyoto3 datasets, respectively.

Deep Learning Approaches for Recognizing Daily Human Activities Using Smart Home Sensors

The development of sensor technology has enabled the development of a variety of applications for human activity detection by wearable devices. Identifying transportation modes for contextual support in the execution of systems such as driver assistance and intelligent transportation planning is one of the advantageous applications of an intelligent transportation system (ITS). Due to the widespread use of smartphones in today’s world, a mobile application-based solution is proposed that can significantly reduce the cost of implementing ITS. In this work, we recognized transportation vehicles using the accelerometer and gyroscope data collected by smartphones. To achieve the research goal, this work developed a deep residual network called DeepResNeXt that used convolutional kernels and residual connections for transportation vehicle recognition. We used a public benchmark dataset to evaluate the proposed deep residual network. Experimental results showed that DeepResNeXt was achieved better accuracy and F1-score than previous works. In addition, this work also investigated the effect of sensor types on recognition performance. The results showed that the deep residual network trained with accelerometers achieved higher accuracy and F1-score than the network trained with gyroscope data.

A Deep Residual Network for Recognizing Transportation Vehicles using Smartphone Sensors

In the latest days, study into the development of intelligent technologies has proven valuable, contributing to attempts to improve the quality of human existence. Smart glass is one of the intelligent wearable devices that can be used for various purposes, including healthcare monitoring, fall detection, sleep tracking, and human activity recognition (HAR). Smart-phones and smartwatches are the primary wearables utilized in sensor-based HAR to collect human motions for training recognition models based on physical movement. These wearable tools, nevertheless, are more intrusive than smart glasses. Using IMU sensor data acquired via smart glasses, we investigate deep learning algorithms for detecting people's activities of daily living (ADL). This work proposes a hybrid deep neural network that automatically extracts spatial-temporal information from raw data to enhance identification $\mathbf{p}$ erformance. We performed tests to evaluate deep learning models using a publically available benchmark dataset, UCA-EHAR, which included IMU sensor data from multiple ADL from smart eyewear. The recommended CNN-LSTM model achieved the best effectiveness with the highest F1-score of 93.20%, as determined by experimental findings.

Deep Learning Approaches for HAR of Daily Living Activities Using IMU Sensors in Smart Glasses

Recognition of human activity has utilized inputs from wearable sensors, which has significant implications for rehabilitative medicine and cognitive neuroscience. Unfortunately, some crucial dynamic data on upper-limb movements need to be included in the feature extraction procedure for wearable sensor data. The issue is that only a few rehabilitative motions can be recognized, and classification precision is readily compromised. We study several convolution neural networks to extract valuable characteristics from multichannel wearable sensor inputs automatically and precisely identify rehabilitation operations. We gathered wearable sensor signal data for six physiotherapy exercises to assess identification effectiveness using the SPARS9x standard rehabilitation dataset. Experiments showed that the PyramidNet18 model had the highest F1-score on the benchmark dataset, 99.15%.

Monitoring System of Wearable Sensor Signal in Rehabilitation Using Efficient Deep Learning Approaches

Falls are the contributing factor to both fatal and nonfatal injuries in the elderly. Therefore, pre-impact fall detection, which identifies a fall before the body collides with the floor, would be essential. Recently, researchers have turned their attention from post-impact fall detection to pre-impact fall detection. Pre-impact fall detection solutions typically use either a threshold-based or machine learning-based approach, although the threshold value would be difficult to accurately determine in threshold-based methods. Moreover, while additional features could sometimes assist in categorizing falls and non-falls more precisely, the estimated determination of the significant features would be too time-intensive, thus using a significant portion of the algorithm’s operating time. In this work, we developed a deep residual network with aggregation transformation called FDSNeXt for a pre-impact fall detection approach employing wearable inertial sensors. The proposed network was introduced to address the limitations of feature extraction, threshold definition, and algorithm complexity. After training on a large-scale motion dataset, the KFall dataset, and straightforward evaluation with standard metrics, the proposed approach identified pre-impact and impact falls with high accuracy of 91.87 and 92.52%, respectively. In addition, we have investigated fall detection’s performances of three state-of-the-art deep learning models such as a convolutional neural network (CNN), a long short-term memory neural network (LSTM), and a hybrid model (CNN-LSTM). The experimental results showed that the proposed FDSNeXt model outperformed these deep learning models (CNN, LSTM, and CNN-LSTM) with significant improvements. 

Sakorn Mekruksavanich

Papers

Deep Learning Approaches for Recognizing Daily Human Activities Using Smart Home Sensors

A Deep Residual Network for Recognizing Transportation Vehicles using Smartphone Sensors

Deep Learning Approaches for HAR of Daily Living Activities Using IMU Sensors in Smart Glasses

Monitoring System of Wearable Sensor Signal in Rehabilitation Using Efficient Deep Learning Approaches

Pre-Impact and Impact Fall Detection Based on a Multimodal Sensor Using a Deep Residual Network