Consiglio Per La Ricerca in Agricoltura E L'Analisi Dell'Economia Agraria (CREA), Centro "Zootecnia E Acquacoltura", Via Salaria 31, 00015, Monterotondo, Rome, Italy.
Sci Rep. 2022 Sep 19;12(1):15642. doi: 10.1038/s41598-022-19932-9.
Common aquaculture practices involve measuring fish biometrics at different growth stages, which is crucial for feeding regime management and for improving farmed fish welfare. Fish measurements are usually carried out manually on individual fish. However, this process is laborious, time-consuming, and stressful to the fish. Therefore, the development of fast, precise, low cost and indirect measurement would be of great interest to the aquaculture sector. In this study, we explore a promising way to take fish measurements in a non-invasive approach through computer vision. Images captured by a stereoscopic camera are used by Artificial Intelligence algorithms in conjunction with computer vision to automatically obtain an accurate estimation of the characteristics of fish, such as body length and weight. We describe the development of a computer vision system for automated recognition of body traits through image processing and linear models for the measurement of fish length and prediction of body weight. The measurements are obtained through a relatively low-cost prototype consisting of a smart buoy equipped with stereo cameras, tested in a commercial mariculture cage in the Mediterranean Sea. Our findings suggest that this method can successfully estimate fish biometric parameters, with a mean error of ± 1.15 cm.
水产养殖的常见做法包括在不同的生长阶段测量鱼类的生物计量学，这对于饲养管理和提高养殖鱼类的福利至关重要。鱼类的测量通常是在个体鱼类上进行手动完成的。然而，这个过程既费力、耗时又对鱼类有压力。因此，开发快速、精确、低成本和间接的测量方法将受到水产养殖部门的极大关注。在这项研究中，我们探索了一种通过计算机视觉对鱼类进行非侵入性测量的有前途的方法。立体相机拍摄的图像由人工智能算法与计算机视觉相结合，自动获得鱼类特征（如体长和体重）的精确估计。我们描述了一种计算机视觉系统的开发，该系统通过图像处理和线性模型进行自动识别身体特征，用于测量鱼类的长度和预测体重。该测量是通过一个相对低成本的原型获得的，该原型由一个配备立体相机的智能浮标组成，在地中海的一个商业海水养殖笼中进行了测试。我们的研究结果表明，该方法可以成功地估计鱼类的生物计量参数，平均误差为±1.15 厘米。
