Object detection in inland vessels using combined trained and pretrained models of YOLO8

Abstract

One of the central challenges within the domain of computer vision is object detection, encompassing the identification and localization of specific entities within an image. Introducing a pioneering approach, the YOLO (You Only Look Once) algorithm emerged in 2015, executing object detection within a singular neural network. This innovation triggered a profound transformation within object detection, ushering in remarkable advancements beyond the capacities of the preceding decade. Subsequently, YOLO underwent successive iterations, culminating in eight versions that have earned prominent stature among leading object identification algorithms. This recognition is attributed to YOLO's integration of state-of-the-art concepts prevalent in the realm of computer vision research. Particularly noteworthy is the latest iteration, YOLOv8, which demonstrates superior performance in terms of both accuracy and speed when juxtaposed with YOLOv7 and YOLOv5. This study delves into the most recent strides in object detection as an important field of computer vision, which has been seamlessly assimilated into YOLOv5, YOLOv7, YOLOv8, and their antecedents. The introductory section, delineating the foundational importance of object detection, aligns seamlessly with the research's overall narrative. The elucidation of object detection's significance within diverse contexts, such as vehicle identification across varying scales and environments, underscores its multifaceted utility. The refinement process further enhances the discernment of YOLO's progression through its iterations, elucidating the evolution from the pre-eminent YOLOv1 to the recent apex represented by YOLOv8. Notably, the text now highlights YOLOv8's distinc-tive advancements in accuracy and speed over YOLOv7 and YOLOv5, lending heightened clarity to the incremental evolution of the algorithm. The augmentation extends to the exploration of YOLOv8's amalgamation with contemporary computer vision concepts. These concepts' incorporation is now underscored, demonstrating how YOLOv8 benefits from the strides made in computer vision research. The final passage captures the thrust of the research, examining the application of the developed object detection models within the specific context of inland waterway vessels. The distinct stages of detection, the addition of new classes, manual annotation, and the process of network training are now presented with greater precision,