基于计算机视觉的3种金枪鱼属鱼类形态指标自动测量研究

金枪鱼类是我国远洋渔业重要的捕捞对象，其形态指标对研究金枪鱼类的生长、发育和生活史具有重要意义。人工测量形态指标是一种非常繁琐且低效率的测量方法，而计算机视觉是一种高效和客观的自动测量方法。因此，本文通过计算机视觉库OpenCV对3种金枪鱼类图像进行预处理，主要利用双边滤波、灰度变换、二值化处理和提取轮廓等图像处理技术得到金枪鱼类形态轮廓图像。根据预先选定的特征点，利用计算机视觉技术遍历轮廓图像上所有的像素点，并自动定位出每张轮廓图像的预选特征点共17个。利用计算机视觉技术遍历得到的特征点位置，自动测量出3种金枪鱼的形态指标像素长度，并计算出形态指标实际长度。还分析自动测量与人工测量形态指标的绝对误差和相对误差。研究结果表明，通过计算机视觉技术对3种金枪鱼的形态指标的自动测量效果较好，大眼金枪鱼、黄鳍金枪鱼和长鳍金枪鱼的12个形态指标的绝对误差范围分别为0～1.46 cm、0～1.73 cm、0～1.32 cm，其相对误差范围分别为0.01%～5.84%、0%～6.17%、0%～6.89%。本研究以期为金枪鱼类智能识别提供前期工作基础，也为其他鱼类自动测量研究提供基础参考。

Automatic measurement of morphological indexes of three Thunnus species based on computer vision

Tuna is an important fishing target in China’s pelagic fishery. Its morphological indexes are of great significance for the study of the growth, development and life history of tunas. Manual measurement of morphological index is a very tedious and inefficient measurement method, while computer vision is an efficient and objective automatic measurement method. Therefore, in this paper, images of three Thunnus species are preprocessed by the computer vision library (OpenCV). It mainly uses image processing techniques such as bilateral filter, gray transformation, image binarization and contour extraction to obtain the contour image of tuna. According to the pre-selected feature points, the computer vision technology is used to traversal all the pixel points on the contour image, and 17 pre-selected feature points of each contour image are automatically located. By using the computer vision technology, the pixel length of the morphological index of the three species of tuna is automatically measured and the actual length of the morphological index is calculated. The absolute error and relative error between automatic measurement and manual measurement are compared and analyzed. The results show that the computer vision technique is effective in the automatic measurement of the morphological indexes of the three Thunnus species. The absolute error ranges of 12 morphological indices of Thunnus obesus, Thunnus albacores and Thunnus alalunga are 0.00?1.46 cm, 0?1.73 cm and 0?1.32 cm, respectively, and the relative error ranges are 0.01%?5.84%, 0%?6.17% and 0%?6.89%, respectively. It is expected to provide a basis for intelligent identification of tuna and a basic reference for automatic measurement of other fish.