【摘要】 针对由不良坐姿引发的身体疾病发生率不断上升的现状，论文综合ShuffleNetV2、注意力机制和长短期记忆网络构建了一种识别不良坐姿的系统。将注意力机制融入ShuffleNetV2的网络结构使调整后的模型可用在对人体姿态的估计；再将得到的人体骨架序列使用长短期记忆网络进行处理得到最终的坐立姿态分类结果。实验结果表明，在自建的单人坐立姿态数据集上该算法表现良好且执行效率高。最后，利用自建的不良坐姿行为数据集，构建了轻量级的不良坐姿识别系统。更多还原

【Abstract】 In view of the current situation where the incidence of physical diseases caused by poor sitting postures is constantly rising,this paper constructs a system for identifying poor sitting postures by integrating ShuffleNetV2,the attention mechanism and the Long Short-Term Memory Network(LSTM). Integrating the attention mechanism into the network structure of ShuffleNetV2enables the adjusted model to be used for the estimation of human postures. Then,the obtained human skeleton sequences are processed by the Long Short-Term Memory Network to obtain the final classification results of sitting postures. The experiment results show that the performance of algorithm is good and efficient on self-structure database of single person poses. By using this database,the light-weight recognition system for bad sitting pose is realized.更多还原