摘要： 为获取金鲳鱼质量分级所需的质量信息，实现快速无接触式预处理加工。基于多条金鲳鱼的正反面投影图像，采用图像处理算法获取投影面积、周长、长短轴等数据，结合人工测量的金鲳鱼体长、体宽、鱼鳍质量、头鳍宽、鱼尾最窄处宽等关联数据，通过线性回归、曲线估算、主成分分析、RBF神经网络等方法，建立质量预测模型并进行验证。结果显示：建立的14种质量预测模型中，基于图像面积、鱼鳍质量占比、去鱼鳍面积占比的多元线性回归模型具有较高的预测精度，模型的决定系数为0.919。基于图像面积、图像周长、长轴、短轴的线性回归模型决定系数为0.918。建立的9种主成分回归模型中二次回归模型最优，其决定系数为0.877，标准估计误差为17.094。通过建模样本外数据对模型进行验证，基于图像面积、鱼鳍质量占比、去鱼鳍面积占比的回归模型预测值与实际值相关系数为0.944，平均相对误差为2.43%，标准差为2.32%，试验结果表明，基于机器视觉技术的金鲳鱼质量预测模型可靠性高，可应用于金鲳鱼质量分级。

关键词: 机器视觉技术, 质量分级, 预测模型, 金鲳鱼

Abstract: To obtain the weight information required for weight classification of Golden pomfret and realize rapid and non-contact preprocessing. This research is based on the front and back projection images of multiple Golden pomfret, and an image processing algorithm is used to obtain the projected area, perimeter, long and short axes, etc. and is combined with artificially measured body length, body width, fin weight, head fin width, the width of the narrowest point of fishtail and other related data, through linear regression, curve estimation, principal component analysis, RBF neural network, and other methods build a weight prediction model and validate it. The results show that among the 14 weight prediction models established, the multiple linear regression model based on the image area, the proportion of fin weight, and the proportion of finless area have higher prediction accuracy, and the model's coefficient of determination is 0.919. The coefficient of determination of the linear regression model based on the image area, image perimeter, long axis, and short axis is 0.918. Among the established 9 principal component regression models, the quadratic regression model is the best, with a coefficient of determination of 0.877 and a standard estimation error of 17.094. The model is verified by modeling out-of-sample data. The correlation coefficient between the predicted value and the actual value of the regression model based on the image area, the proportion of fin weight, and the proportion of the finless area is 0.944, the average relative error is 2.43%, and the standard deviation is 2.32%. The research results show that the quality prediction model of Golden pomfret based on machine vision technology has high reliability and can be applied to the quality classification of Golden pomfret.

Key words:  , machine vision technology, weight classification, prediction model, Golden pomfret