Toward Improving the Performance of Hand Movement Intention Recognition for Upper Limb Amputees Using Flexible Ultrasound Transducers

In recent years, with the development of ultrasonic sensing technology, ultrasound is considered as a promising alternative to surface electromyography for the recognition of human hand movements. However, the existing ultrasound transducer is bulky and poor wearability, and the performance of ultrasound based motion recognition for amputees was not comprehensively investigated yet. So, in this study, we used a wearable, flexible and lightweight B-mode ultrasound probe to detect the morphological change of muscles of 24 fine hand movements for four subjects. In order to develop an effective B-mode ultrasound based motion recognition method, the performances of five kinds of image features and three kinds of classification algorithms were systematically investigated. And the results showed that the combination of HOG+SVM had the highest average recognition accuracy of 97.84% and the combination of Sobel+SVM was the most robust to individual difference. The work of this study may provide help for the application of muscle ultrasound image based motion recognition.