基于机器视觉的水下鲆鲽鱼类质量估计

Weight estimation of underwater Cynoglossus semilaevis based on machine vision

为了更好地解决水下鲆鲽类等底栖鱼的质量估计问题,本研究获取了半滑舌鳎(Cynoglossus semilaevis)在不同生长阶段的图像和质量数据,利用图像处理技术测量出鱼的面积,并将面积与质量进行数据拟合建立模型。结果表明面积与质量的相关性可达到0.9682,测试平均相对误差为6.17%。由于鱼的质量还受其他形状参数的影响,同时测量了等效椭圆长短轴比和圆形度因子,对面积、等效椭圆长短轴比和质量,面积、圆形度因子和质量分别进行三维拟合,质量估计的平均相对误差分别为5.50%、5.62%。通过验证表明,对水底鱼拍摄的图像经过水底模板校准后的处理结果,与水外面拍摄处理后的结果一致,因此在不捕捞的情况下可以实现水底活体鲆鲽类鱼的质量估计。

In order to solve the problem of estimating the weight of the flounder and other flatfishes underwater, the image and weight data of the tonguefish （Cynoglossus semilaevis） were obtained at different growth stages. The area of the fish were measured through the image processing. The fitting models were established to match the area comparaed with weight. The results showed that the correlation between the area and the weight could reach at 0.9682, the average relative error was 6.17%. In addition, the weight was also affected by other shape parameters, and the ration of equivalent ellipse axes and heywood circularity factor were measured. Three-dimensional fitting models of area, ration of equivalent ellipse axes, weight, and area, heywood circularity factor, weight were established. The average relative errors of the two models were 5.50%, 5.62%, respectively. The experiment verified that the processed results of images of fish underwater obtained out of the surface and calibrated with the template underwater were consistent with that of images obtained outside the water. Consequently, the weight of flounder fish underwater can be estimated without catching the fish.