نوع مقاله : مقاله کامل پژوهشی
نویسندگان
1 دانشجوی کارشناسی ارشد، گروه الکترونیک، دانشکدهی مهندسی برق، دانشگاه علم و صنعت ایران، تهران، ایران
2 استاد، گروه الکترونیک، دانشکدهی مهندسی برق، دانشگاه علم و صنعت ایران، تهران، ایران
کلیدواژهها
موضوعات
عنوان مقاله English
نویسندگان English
Examining the sound of body organs is one of the methods of diagnosing various diseases, which is used by specialists to analyze abnormal sounds. Since most of the deaths caused by the disease occur in poor countries that have a shortage of equipment and specialists, the development of diagnostic methods based on machine learning and audio processing, which are available, non-invasive and inexpensive, can lead to early diagnosis and save the lives of millions of people. In previous studies, inputs that reflect the frequency characteristics of the sound have been used, in this article, we also use a recurrent representation that reflects the temporal characteristics of the sound and is given as an input to convolutional networks in order to benefit from its transfer learning advantages. By adding the temporal attention mechanism and the bidirectional recurrent gates, the audio data sequence which is a time series is investigated and each data is weighted according to its value. The data used in this article is from the ICBHI lung sound database, which has been used in many other articles. The presented method was able to classify lung sounds into three categories: healthy, chronic obstructive pulmonary disease (COPD) and other diseases with an accuracy of 97%, which is a better result than other methods that used this database.
کلیدواژهها English