As an advanced manufacturing technique, high-energy abrasive water jet (AWJ) has some special advantages such as high efficiency, no-heat affected zone, cleanliness. Therefore, it is particularly suitable for processing some special materials used extensively in aerospace, rail, and shipbuilding industries. However, the jet undergoes a real-time dynamic deformation during cutting process due to the changes of the cutting conditions. And the dynamically deforming jet often results in low cutting accuracy. If the cutting front profile of the jet can be accurately predicted and controlled properly, machining precision can be greatly enhanced. In this study, a high-speed camera has been utilized to capture the images of the jet during the jet cutting process. After that, a dataset of jet image segmentation has been created, and the jet profiles have been extracted from each image using the Segment Anything Model (SAM). We represented and quantified the trailing edge features of the jet profiles and proposed an AWJ_GA_BP neural network-based jet profile prediction algorithm, achieving a more accurate prediction of trailing edge features of the jet profiles. This research not only improves the understanding of abrasive water jet machining mechanisms but also provides a crucial guidance for precise control of abrasive water jet.