Ornamental fish dataset from an underwater environment

In underwater environments, there are more than 33,000 species of fish, which are identified by different visual characteristics, such as the shape, color, and shape of the head. These characteristics are difficult to identify for ordinary people, therefore scientists and aquaculturists are using photographic and video cameras as tools to quantify the species and identify the state and shape. Providing these tools with computer vision algorithms and deep learning techniques, since recorded images of fish can be time-consuming and expensive to process and analyze manually, it is an interesting problem for researchers. However, the use of these techniques depends on the visual characteristics extracted by the set of images with which it has been trained. This article presents an image data set of 3 different fish species, Goldenfish (Carassius auratus), Molly (Poecilia sphenops), and Zebra (Danio rerio), obtaining a different number of images per species. The data were obtained by means of a camera in a natural environment where the species, housed in the environment protected by the aquaculturist.

水下环境中现存超过33000种鱼类，可通过外形、体色及头部形态等多种视觉特征进行区分。这类特征普通大众难以辨识，因此科研人员与水产养殖从业者借助摄影及摄像设备，实现鱼类物种的量化统计以及个体状态与形态的识别。由于手动处理与分析鱼类录制影像既耗时又成本高昂，为这类工具集成计算机视觉（Computer Vision）算法与深度学习（Deep Learning）技术，便成为研究者们颇具研究价值的课题。不过此类技术的应用效果，依赖于训练模型所用图像集所提取的视觉特征。本文构建了涵盖3种不同鱼类的图像数据集，分别为金鱼（Carassius auratus）、摩利鱼（Poecilia sphenops）与斑马鱼（Danio rerio），各物种对应的图像数量各不相同。本数据集通过摄像机在自然环境中采集，该环境由水产养殖者搭建防护设施，供目标鱼类栖息。

搜集汇总

背景与挑战

背景概述

该数据集是一个水下环境观赏鱼图像数据集，包含Goldenfish、Molly和Zebra三种鱼类的图像，每物种图像数量不等，旨在支持计算机视觉和深度学习研究，以自动化识别鱼类物种和状态。数据通过自然环境下相机拍摄获取，突出了其真实应用场景，解决了人工分析水下鱼类图像耗时昂贵的问题。