YOLOv7 and ODM

In this particular investigation, the YOLOv7 [23] method is used for fish object identification. This decision was taken after some time was spent previously weighing the benefits of several YOLO series algorithms. To identify fish targets, we implemented several modifications to YOLOv7, one of which was the complete replacement of the feature extraction network as well as an upgrade to the existing feature extraction network [80]. These changes were done to classify nine different types of fish species. YOLOv7 is comprised of three fundamental components: the Backbone network, which is in charge of feature extraction; the improved feature extraction networks; and the YoloHead network, which is in charge of prediction [81]. The original YOLO network has been improved with the introduction of the YOLOv7 platform. There have been enhancements made to several different aspects, such as the feature extraction network, the activation function, the loss function, and several other areas. Both the activation function, which is changed from Leaky ReLU to Mish, and the network for feature extraction, which is changed from YOLOv7’s [82] Darknet53 to CSP Darknet53, are updated to reduce the size of the model while maintaining the same level of accuracy. The version of the YOLO algorithm known as YOLOv7 is both more accurate and more productive when compared to previous iterations of the YOLO algorithm [83].