Gesture recognition plays a vital role in complex human-machine interaction maintenance scenarios and serves as a key technology for achieving intelligent human-machine collaboration and enhancing operational efficiency. However, conventional deep learning-based gesture recognition methods often suffer from complex network structures and high computational overhead, which hinder their applicability in industrial environments where real-time performance and lightweight deployment are critical—especially in scenes with complex and dynamic backgrounds. To address these challenges, this paper proposes an improved YOLOv8-based gesture recognition framework tailored for industrial applications. The proposed method replaces standard convolutional layers with MSDConv modules to reduce model parameters, integrates the LightContext attention mechanism to enhance focus on key gesture regions while improving robustness in complex background conditions, and employs the WiseIoU loss function to improve detection accuracy. These improvements work synergistically to enhance detection performance while maintaining a lightweight and efficient architecture. Extensive experiments demonstrate the effectiveness of each component improvement. The optimized model achieves a 2.3% increase in recognition accuracy (reaching 96.3%), and improves detection speed by 6%. Furthermore, it enables real-time recognition of 10 gestures in complex background environments, better satisfying the demands for both lightweight design and real-time performance in practical industrial settings.