Hypertension is a major cause of premature death worldwide, and a 5 mmHg error in blood pressure values doubles or halves the number of hypertensive patients. Thus, the accurate measurement of blood pressure is crucial. To date, an auscultatory method where a clinician hears sounds called Korotkoff sounds has been regarded as a gold standard of non-invasive blood pressure measurement. In this study, to measure the blood pressure based on the auscultatory method without the help of a clinician, the Korotkoff sound signal is converted into featured images using wavelet transforms. The featured images are used as inputs of the convolutional neural network, and the network is trained to classify the images that correspond to the valid Korotkoff sound. The classification result showed that the valid Korotkoff sound was detected with an accuracy of 91% on average.