Research on Ear Recognition Based on SSD_MobileNet_v1 Network
Yanmin Lei, Baowei DU, Junru Qian, Zhibin Feng · 2020
Human ear recognition technology is an emerging technology in biometrics. It has high theoretical research value and market application prospects, and it gradually develops with the development of image processing, pattern recognition and other fields. Aiming at the problem of insufficient accuracy and detection speed of human ear image recognition in practical applications, This paper constructs a human ear image recognition method based on SSD_MobileNet_v1 target detection model. The data set uses the USTB human ear image library of University of Science and Technology Beijing, and the image is enhanced and annotated to eliminate the influence of noise on recognition. Under the TensorFlow platform, MobileNet is used to extract features, and feature regions are generated through RPN region suggestions. And input this characteristic area into the SSD network for training, and apply the trained SSD_MobileNet_v1 model to classify the test image and get the recognition result. Experimental results show that the construction method is accurate in target location, and the recognition accuracy is over 99%, and it has good robustness to images with background interference.