【摘要】 基于深度学习的坐姿识别方法目前在识别精度上获得了较好提升，但模型在嵌入式平台上无法兼具高准确性和快速性，进而难以应用于边缘智能等领域。基于神经网络提出了一种坐姿快速识别方法，使用轻量化网络替换骨干网络提取底层特征，并利用基于自适应批量归一化层候选评估模块对模型进行剪枝优化。同时，对坐姿识别方法进行改进，在骨骼关节特征的基础上融合骨骼图像特征，在提升检测速度的同时保证了识别精度。实验结果显示，改进后的模型识别准确率高，并且检测速度获得大幅提升。更多还原

【Abstract】 The accuracy of sitting posture recognition based on deep learning has been improved, but the model cannot possess the characteristics of both high accuracy and rapidity at the same time on the embedded platform, and it is difficult to be applied to fields such as edge intelligence. A fast sitting posture recognition method based on neural network is proposed. Light weighted network is used to replace backbone network to extract underlying features, and adaptive batch-based normalized layer candidate evaluation module is used to optimize the model. At the same time, the sitting posture recognition method is improved, and the bone image features are merged on the basis of the bone joint characteristics, which improves the detection speed while ensuring the recognition accuracy. The experimental results show that the accuracy rate of the improved model is high and the detection speed has also been greatly improved.更多还原