Data underlying the publication: Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review

Data for training and evaluation of a method for detection and counting demersal fish species in complex, cluttered and occluded environments that can be installed on the conveyor belts of fishing vessels. The data mainly exists of images of fish on a conveyer belt with the corresponding annotations. This was used to train a neural network (YOLOv3) to detect and classify fish species. Because each fish is visible in multiple images, the fishes were tracked over consecutive images and the total number of fish per specie was counted. These counts were compared to human review.

本数据集用于训练和评估一种可安装于渔船传送带之上、面向复杂杂乱且存在遮挡环境中底栖鱼类物种检测与计数的方法。该数据集主要由传送带上的鱼类图像及其对应标注信息构成，曾被用于训练神经网络YOLOv3以实现鱼类物种的检测与分类。由于单条鱼类会在多张连续图像中出现，研究人员对各鱼类在连续帧间进行了追踪，并统计得到每个物种的总个体数，随后将上述统计结果与人工评审结果进行对比验证。