计算机科学
计算
声音(地理)
领域(数学分析)
频域
人工智能
语音识别
算法
声学
数学
计算机视觉
数学分析
物理
作者
Rui Zhang,Xinyu Li,Lei Pan,Jing Hu,Peng-Yun Zhang
标识
DOI:10.1016/j.bspc.2024.106332
摘要
This paper proposes a heart sound diagnosis method based on multi-domain self-learning convolutional computation. It aims to address the limitations of the existing heart sound time-domain samples' own feature expression, the inadequacy of model feature mining and the subjectivity of manual parameter tuning. A multi-domain mapping processing method is used to further expand the self-feature expression ability of heart sound samples. A heart sound feature mining model that fits the multi-domain samples is designed to further mine disease features from different receptive fields and multi-resolutions in each spatial domain. Collaborative computation is employed to enhance the richness of feature mining. Finally, a model optimization strategy based on the parallel sand cat swarm optimization algorithm is constructed to optimize the 12 key hyperparameters of the model, so that the model can more adequately extract the sample discriminative features ultimately realizing the efficient diagnosis of heart sounds. The self-optimization diagnostic model constructed in this paper achieved a diagnostic accuracy of over 99 % in the experiments conducted on two public data sets, Yaseen (2018) and the Heart Sound Classification Challenge B. These results show an improvement of at least 1.27 % and 2.61 % compared to other research methods mentioned in the literature. The proposed method in this paper is feasible and effective, providing theoretical and technical references for the field of heart sound diagnosis and automatic machine learning. It holds certain scientific research value.
科研通智能强力驱动
Strongly Powered by AbleSci AI