Zeng Mengjun

Papers

Total Citations

H-Index

About

Zeng Mengjun has made significant contributions to the intersection of deep learning and robotics, with a primary focus on keyword spotting (KWS) and object recognition for humanoid platforms. His most influential work, "Effective Combination of DenseNet and BiLSTM for Keyword Spotting" (2019, 83 citations), pioneered a novel architecture that fuses DenseNet’s powerful local feature extraction with BiLSTM’s sequential modeling capabilities. This approach dramatically improved detection accuracy while maintaining a compact footprint—a critical requirement for smart on-device terminals and service robots. By addressing the core challenge of maximizing performance within tight computational constraints, Zeng’s research has directly advanced the practicality of voice-controlled human-robot interaction. In his subsequent work, "Optimized Convolutional Neural Network-Based Object Recognition for Humanoid Robot" (2020, 4 citations), he extended his expertise to visual perception, designing optimized CNN architectures tailored for real-time object recognition in anthropomorphic robots. Zeng’s research bridges the gap between theoretical deep learning advances and real-world robotic applications, demonstrating how careful architectural design can enable more capable, responsive, and autonomous humanoid systems. His work continues to inspire researchers seeking efficient, deployable AI solutions for interactive robotics.