In marine aquaculture, fish populations constantly decrease throughout the cultivation period because of mortality and escape. Current production management systems provide limited opportunities to count the cultured fish, making it difficult to estimate accurately the fish population in the cage. To overcome this problem, an automatic fish identifying method based on particle tracking velocimetry (PTV) flow visualization technology is proposed in this paper. The proposed method utilizes an image processing unit that extracts individual fish from the acquired image and a motion analysis unit that calculates the motion vector for each individual. Thus, the accuracy of the extraction results in the image processing unit affects the system’s counting results. To validate the efficiency and robustness of the image extraction performed by the image processing unit, individuals were extracted from images using the open-source image deep learning semantic segmentation method (SegNet), which is able to distinguish between the background and foreground in the images via analysis at the pixel level. SegNet is able to improve the image discrimination performance by multiplying the learning paths, and the robustness of the detection results can be ensured by changing the layer structure according to the detection target. Accordingly, the use of SegNet was evaluated in terms of the number of layers and images in the training set. The results of this study indicate that the application of SegNet with PTV technology represents a promising method for the automatic identifying and behavioral tracking of fish in an aquaculture net cage.

在海水养殖中，由于死亡和逃逸，整个养殖期间鱼类种群不断减少。当前的生产管理系统为养殖鱼的计数提供了有限的机会，这使得很难准确估算网箱中的鱼类数量。为了克服这个问题，本文提出了一种基于粒子跟踪测速流可视化技术的鱼类自动识别方法。所提出的方法利用从所获取的图像中提取单个鱼的图像处理单元和为每个个体计算运动矢量的运动分析单元。因此，图像处理单元中提取结果的准确性会影响系统的计数结果。为了验证图像处理单元执行的图像提取的效率和鲁棒性，使用开源图像深度学习语义分割方法（SegNet）从图像中提取了个人，该方法能够区分图像中的背景和前景通过像素级别的分析。SegNet可以通过增加学习路径来提高图像识别性能，并且可以根据检测目标更改层结构来确保检测结果的鲁棒性。因此，根据训练集中的层数和图像评估了SegNet的使用。