With the development of modern cars, the auto-motive Ethernet has become a backbone network of the in-vehicle network in recent years. The open interfaces of the automotive Ethernet increase risk of being attacked. Device fingerprints identification can be used for the access authentication by utilizing the hardware characteristics of the devices. The signals are intertwined with random data from the both sides of communication which improves the difficulty of feature extraction in Ethernet devices. We propose a feature extraction method based on power spectrum for the 100BASE-T1, in which the feature can be utilized for the coupled signals and remove the influence of the random data. We built the automotive Ethernet device identification platform based on 100Base-T1 with 10 devices to be identified. According to the test, the recognition rate is above 95% when we take the average of 1000 segments' power spectrum with 1024 sample points.