Understanding the breaking characteristics of waves is important in several nearshore applications such as assessing impacts of submerged engineered structures on wave breaking or computing surf zone energy budgets. Past studies have used images collected by remote sensing to estimate characteristics such as breaking wave height, depth, position, and type (e.g., plunging, spilling, non-breaking). Due to the dynamic nature of breaking waves, breaker classification from a single image may have large uncertainty. For this reason, an approach involving multiple frames is explored. Here, we develop a you only look once – random forest (YOLO-RF) machine learning (ML) model to predict breaker type (plunging or spilling) from GoPro video data shot cross-shore at oncoming waves (face-on). A YOLO model which classifies five wave features (i.e., prebreaking, curling, splashing, whitewash, crumbling) in a set of video frames is coupled to an RF model which takes normalized feature counts over multiple frames as inputs, and outputs a wave-breaking type for each detected wave. The YOLO model detects wave features as separate objects allowing for individual classification of waves in the same frame. The model, trained and validated with data from a large-scale wave-flume experiment, identifies breaker type with 94 % accuracy proving useful for consistent laboratory data. Only a small subset of cases needs to be labeled by hand for training, while the remainder can be labeled by the YOLO-RF model. This open-source approach could be adapted for field settings to aid in understanding, predicting, and modeling wave breaking dynamics in the nearshore environment.
扫码关注我们
求助内容:
应助结果提醒方式:
