Article: Basketball player action optimisation based on deep reinforcement learning: multimodal biomechanical modelling Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.51 pp.1 - 17 Abstract: Basketball player action optimisation is key to improving competitive performance. To address the issue of insufficient mining of biomechanical features and poor action optimisation effects in current research, this paper first conducts biomechanical data analysis of basketball players and constructs a motion state equation for capturing and analysing key data. Then, it integrates image and biomechanical motion data to provide comprehensive multimodal perception information. A multimodal feature extraction and fusion module based on self-attention mechanism is designed. Secondly, the pose action decision-making task of the athlete is modelled as a deep reinforcement learning (DRL) problem. Finally, a hybrid reward function is designed to achieve efficient training of the model and action strategy optimisation. Experimental outcome indicates that the high model improves the action optimisation success rate by at least 5% compared to the baseline model, demonstrating good action optimisation effects. Inderscience Publishers - linking academia, business and industry through research

Title: Basketball player action optimisation based on deep reinforcement learning: multimodal biomechanical modelling

Authors: Tao Huang

Addresses: School of Physical Education, Huanggang Normal University, Huanggang, 438000, China

Abstract: Basketball player action optimisation is key to improving competitive performance. To address the issue of insufficient mining of biomechanical features and poor action optimisation effects in current research, this paper first conducts biomechanical data analysis of basketball players and constructs a motion state equation for capturing and analysing key data. Then, it integrates image and biomechanical motion data to provide comprehensive multimodal perception information. A multimodal feature extraction and fusion module based on self-attention mechanism is designed. Secondly, the pose action decision-making task of the athlete is modelled as a deep reinforcement learning (DRL) problem. Finally, a hybrid reward function is designed to achieve efficient training of the model and action strategy optimisation. Experimental outcome indicates that the high model improves the action optimisation success rate by at least 5% compared to the baseline model, demonstrating good action optimisation effects.

Keywords: basketball action optimisation; deep reinforcement learning; biomechanical modelling; multimodal feature fusion; attention mechanism.

DOI: 10.1504/IJICT.2025.151074

International Journal of Information and Communication Technology, 2025 Vol.26 No.51, pp.1 - 17

Received: 17 Jul 2025
Accepted: 08 Sep 2025
Published online: 12 Jan 2026 *

Title: Basketball player action optimisation based on deep reinforcement learning: multimodal biomechanical modelling

Keep up-to-date