Khail-Net: A Shallow Convolutional Neural Network for Recognizing Sports Activities Using Wearable Inertial Sensors
Hamza Ali Imran
Abstract
Time-series data can be generated by wearable sensors such as accelerometers, gyroscopes, and magnetometers. This data may be used to classify various everyday life activities using machine learning or deep learning models. Athletics, education, child monitoring, ambient assisted living, and other applications benefit from human activity recognition. Human activity recognition includes sports activity recognition. A typical sports activity is any action that is often employed in a variety of sports. Walking, jogging, sprinting, and leaping are basic sports motions. A unique sports action exists exclusively in one sport, such as a badminton smash. We proposed a shallow convolution neural network for sports activity recognition. It has just 1251 trainable parameters. The test accuracy attained is 98.387%. The average F1 score, recall, and precision are 98.0, 98.7, and 98.0%, respectively. We have also trained the model on a benchmark human activity recognition dataset from WISDM lab for performance evaluation and comparison of the model.