Muscle fatigue occurs after sports activities, repeated actions in a routine job, or a heavy-duty job. It causes soreness and reduces performance in athletes and workers. Various therapies have been developed to reduce muscle fatigue. Vibration therapy has been used to reduce muscle fatigue and delay muscle soreness. However, its effectiveness remains unclear. Ultrasound images provide a non-invasive diagnosis and instant visual examinations. However, it requires extensive training to analyze ultrasound images. The purpose of this study was to develop an automated classification system of ultrasound images using deep learning to assist clinical diagnosis. The ultrasound images of the biceps muscle were measured from four healthy people. The primary objective of the study was to use the convolutional neural network (CNN) models to classify between the vibration control condition (0 Hz) and vibration test conditions (5, 35, and 50 Hz) with subjects in different time duration the pattern (2 and 10-min). These images were preprocessed to resize to 224 × 224 pixels and augmentation to feed into the dataset, including the augmentation training dataset (74%), validation dataset (15%), and non-augmentation test dataset (11%). This study used the AlexNet, VGG-16, and VGG-19 of CNN models for recognition and classification ultrasound images. These models compared the differences of ultrasound images of biceps after various vibration between two conditions. The results showed that AlexNet has the best performance with the accuracy 82.5%, sensitivity 67.3%, and specificity 99.5% when 10-min 35 Hz local vibration was applied. The deep learning method, AlexNet, shows the potential for automated classification of biceps ultrasound images for assessing treatment outcomes of vibration therapy.