基于改进残差网络和注意机制的生化分析仪异常取样压力分类

Classification of Abnormal Sampling Pressure Based on Improved Residual Network and Attention Mechanism in Biochemical Analyzers

摘要: 取样系统在生化分析仪中一直起着至关重要的作用。如何有效识别取样过程中的异常信号，提高取样精度和准确性是生化分析仪领域面临的重大挑战。本文提出了一种结合改进残差网络和注意机制的分类模型，以准确识别异常取样压力。首先，利用基于改进残差结构的卷积神经网络提取信号的局部特征；其次，结合双向长短期记忆网络，增强网络对时间特征的捕捉能力；最后，引入高效的加法注意机制，为每个提取的特征分配区域判别权重，帮助模型在训练过程中更加重视重要特征，从而提高分类精度。实验是基于从多个全自动生化分析仪测得的不同浓度样品所建立的数据集进行的。该模型在数据集上的总体准确率为96.02%，有望应用于生化检测仪器领域。

Abstract: Sampling systems have always played a crucial role in biochemical analyzers. How to effectively identify abnormal signals in the sampling process and improve sampling precision and accuracy is a major challenge in the field of biochemical analyzers. In this paper, a classification model combining improved residual network and attention mechanism is proposed to accurately identify abnormal sampling pressure. Firstly, a convolutional neural network based on the improved residual structure is used to extract local features of the signal; secondly, a bidirectional long and short-term memory network is combined to enhance the network’s ability to capture temporal features; finally, an efficient additive attentional mechanism is introduced to assign regional discriminative weights to each extracted feature, which helps the model to pay more attention to the important features during the training process, thus improving the classification accuracy. The experiments are based on a dataset built from samples with different concentrations measured from several fully automated biochemical analyzers. The model has an overall accuracy of 96.02% on the dataset and is expected to be applied in the field of biochemical testing instruments.