An improved black-winged kite algorithm optimized back-propagation neural network for biceps curl classification

Abstract

Accurately identifying and classifying biceps curl types is of vital importance for sports training and upper limb joint rehabilitation training. It can improve the effect and reduce the risk of injury caused by incorrect training. In this study, a dataset of biceps curl training was obtained by measuring wearable sensors. After data preprocessing, 340 samples of 35-dimensional feature data were obtained. The classification labels of the dataset were marked as 1-5 according to the five types of biceps curl. This study proposed a black-winged kite algorithm (IBKA) that uses the good point set (GPS) method and the adaptive spiral search rule, a multi-strategy. IBKA optimized the initial weights, biases, and hidden layer numbers and provided them to the back-propagation neural network (BPNN) to establish the IBKA-BPNN model. The constructed IBKA-BPNN model improved the classification accuracy of the training set from 79.83% to 94.54%, and the accuracy of the test set from 69.61% to 88.33%. The IBKA-BPNN model proposed in this study provides a reliable decision-making basis for real-time coaching, athlete performance analysis, and upper limb rehabilitation. Future work will expand the dataset, integrate more bio signals, and explore lightweight deployment on wearable hardware.