Abstract: In response to the low adoption rate of intelligent agricultural machinery among small and medium-sized farmers and the insufficient adaptability of traditional interaction systems, this paper develops a multimodal intelligent interaction system tailored for complex field scenarios. By integrating noise-resistant speech recognition, millimeter-wave radar gesture control, and anti-glare touch technology, a collaborative interaction system is constructed with speech as the primary mode, gesture as the auxiliary mode, and touchscreen for fine-tuning. This approach addresses the issues of complex operation, high false-touch rates, and poor environmental adaptability in traditional systems. The paper innovatively introduces an automotive controller area network (CAN) bus architecture to achieve efficient multimodal data fusion and cross-layer scheduling, thereby enhancing the reliability of command transmission. Based on a modular layered architecture that integrates multi-source sensors, adaptive switching of interaction modes is realized through a dynamic priority arbitration algorithm. The system demonstrates superior stability in scenarios such as crop lodging and high-noise environments, significantly improving human-machine collaboration efficiency and providing a technical paradigm for the intelligent transformation of small and medium-sized agricultural operations. The research results hold significant practical value for promoting the standardization of agricultural machinery interaction and the popularization of intelligent equipment.

Key words: multimodal interaction; intelligent agricultural machinery; speech recognition; gesture control

摘要： 针对中小农户智能化农机普及率低、传统交互系统适应性不足的问题，文章开发了面 向复杂田间场景的多模态智能交互系统。融合抗噪声语音识别、毫米波雷达手势控制及防眩 光触控技术，构建语音主导、手势辅助与触屏微调的协同交互体系，解决传统系统操作复杂、 误触率高和环境适应性差的问题。文章创新性地引入汽车电子控制器局域网（CAN）总线架 构，实现多模态数据高效融合与跨层调度，提升指令传输可靠性。基于模块化分层架构集成 多源传感器，通过动态优先级仲裁算法实现交互模式自适应切换。系统在作物倒伏、强噪声 等场景中表现出优越的稳定性，显著提升人机协作效率，为中小农业智能化转型提供技术范 式。研究成果对推动农机交互标准化及智能装备普惠化具有重要实践价值。

关键词: 多模态交互；智能农机；语音识别；手势控制