基于元辅助任务学习的中药饮片识别方法

Traditional Chinese Medicine Slice Recognition Method Based on Meta-assisted Task Learning

中药饮片的分类对临床中药的实际应用有着十分重要的影响,传统的人工分类主要依靠主观经验作为判断依据,而基于计算机视觉的中药饮片自动识别分类有着快速、准确的特点。但影响自动识别结果的因素较多,针对中药饮片自动识别结果受产地、锻造方式等因素影响大的问题,提出了一种基于元辅助任务学习的中药饮片识别方法。该方法采用了辅助任务以提升主任务表现的策略,利用中药饮片的多种属性构成辅助任务,以提升主任务即中药饮片分类结果的准确性;此外该方法还加入了元学习标签生成网络,该网络自动为模型生成辅助标签作为辅助任务,在提升模型表现的同时节省了人工标注的成本;最后该方法使用了Swin-Transformer作为骨干网络进行特征提取,提升了模型的全局感知能力,进一步提升了模型的泛化性。实验结果表明,该方法在不同批次中药饮片中的识别精度均高于普通方法。

The classification of TCM(Traditional Chinese Medicine)decoction pieces has a very important influence on the practical application of clinical TCM.The traditional manual classification mainly relies on subjective experience as the judgment basis,while the automatic recognition and classification of TCM decoction pieces based on computer vision has the characteristics of fast and accurate.However,there are many factors affecting the results of automatic recognition of TCM decoction pieces.Aiming at the problem that the results of automatic recognition of TCM decoction pieces are greatly affected by factors such as origin and forging mode,a method of TCM decoction pieces recognition based on meta-assisted task learning is proposed.The method adopts the strategy of using auxiliary tasks to improve the performance of the main task.Various attributes of TCM decoction pieces are used to form auxiliary tasks to improve the accuracy of the main task,namely,the classification results of TCM decoction pieces.In addition,a meta-learning tag generation network is added to the method.The network automatically generates auxiliary tags for the model as auxiliary tasks,which can improve the performance of the model and save the cost of manual labeling.Finally,the Swin-Transformer is used as the backbone network for feature extraction,which improves the global perception ability of the model and further improves the generalization of the model.The experimental results show that the accuracy of the proposed method in different batches of TCM decoction pieces is higher than that of ordinary method.