Acoustic telemetry is routinely employed to detect and quantify behaviors of aquatic animals. The ability to acoustically tag and release fish and aquatic animals allows researchers to monitor their presence/absence and fine-scale spatial movement. One method of identifying individual tags employs a period signal encoding (HTI signal) method, which utilizes the full transmitted acoustic energy for both detection and identification, providing significantly greater detection ranges than encoded BPSK signal types. However, identifying period encoded tags typically requires manual data review by trained human analysts, resulting in increased processing cost and a lack of scalability. Innovasea conducted a project to determine the efficacy of applying modern Deep Learning techniques to automatically identify animals implanted with period encoded acoustic tags. Specifically, the well-known U-Net model was adapted and trained to detect period encoded acoustic tag signals from raw acoustic telemetry data. The raw data is first transformed into a spectrogram-like image representation and the U-Net learns an image mask that separates actual tag detections from noise and interference. The trained U-Net achieved greater than 95% acoustic tag identification accuracy when compared to tag identification performed by human analysts on the same test data. Numerous examples of the U-Net performance will be presented.