Abstract
Marine mammals have the potential to be harassed by exposure to loud sounds caused by survey or construction activity and even killed by collisions with fast moving vessels. Thus, vessels supporting offshore wind projects must maintain an effective lookout for large cetaceans to prevent harm and comply with relevant Federal regulations. Currently, trained human protected species observers (PSOs) maintain a lookout using binoculars or the naked eye, but this approach requires many trained personnel to maintain a constant lookout and only works during daylight hours. Recent advances in deep learning algorithms, processing hardware, and training data availability have made the prospect of automated marine mammal lookout systems more feasible. Furthermore, thermal infrared (IR) cameras offer the possibility of maintaining a reliable lookout 24-hours per day. Whale blows are often the most prominent feature of a whale on the surface due to their large vertical profiles, and IR cameras can also take advantage of the blow’s high thermal contrast to improve the chances of detection. However, before these automated detection systems can be widely adopted, their efficacy must be proven, which requires numerous test samples in varied conditions. While real-world data collection can provide some instances of whale blows, the exact range to the blow remains uncertain, and replicating the blow for effective and statistically-significant testing across conditions is infeasible. In this paper, we describe the design of an artificial whale blow and a camera-based whale blow detection system, as well as initial results of testing the detector on the artificial blow. In particular, we demonstrate the feasibility of blow detection in IR and begin the process of characterizing detection performance as a function of weather conditions, sea state, and range.