A lightweight CNN heart sound classification method based on multi feature attention

Heart disease is one of the major global health threats, and early diagnosis is crucial for treatment and prevention. Heart sound signals are important indicators for evaluating the health status of the heart, and heart disease classification research is based on heart sound signals. This article proposes a lightweight convolutional neural network (CNN) model based on multi feature fusion and attention mechanism for automatic classification of heart sound signals. This model integrates three features: Discrete Wavelet Transform (DWT), Short Time Fourier Transform (STFT), and Mel Frequency Cepstral Coefficients (MFCC), comprehensively capturing the time-frequency characteristics of heart sound signals. In addition, the channel attention mechanism in the model improves the efficiency and robustness of feature extraction. The experimental results showed that the model achieved a classification accuracy of 99.67% on five types of heart sound signal databases (1 normal and 4 abnormal), with a sensitivity and specificity of approximately 1. Due to its low computational requirements, this model is suitable for real-time deployment in resource limited environments such as portable devices or mobile applications, providing an effective and feasible method for automatic analysis of heart sound signals and early diagnosis of cardiovascular diseases.