基于颜色传感器和遗传算法的牛肉系水力快速检测

为了实现快速检测牛肉系水力,解决肉制品系水力检测不便的问题,该文研究了基于颜色传感器和遗传算法的牛肉系水力快速检测方法。优化了变色试纸与牛肉样品的最佳贴附时间,使用Arduino控制器操控颜色传感器采集试纸颜色参数;将80组牛肉样品分成60个训练组和20个验证组,应用遗传算法优化BP神经网络模型。研究结果表明:在系水力试纸的尺寸为5 cm×2 cm、氯化钴浸泡液浓度为3 g/mL的条件下,得到试纸最佳贴附时间为20 s;经遗传算法优化BP神经网络后,最佳线性回归方程的斜率为0.96,相关系数R~2为0.987,优化后的神经网络模型对系水力等级的预测准确率从90%提高到95%;将检测时间降低到1 min之内,实现了对牛肉系水力等级的快速检测。该研究为进一步开发智能化的肉制品系水力快速检测设备提供了理论依据。

Rapid detection of water-holding capacity in beef using color senor and genetic algorithm

Abstract: In order to achieve the fast and efficient detection of water-holding capacity in meat products, this paper developed a rapid method to detect water-holding capacity in beef using color sensor and genetic algorithm. First of all, the beef was cut into samples with size of 4 cm × 1 cm × 0.5 cm. Then test paper, decorated by cobalt chloride, could change its color when it attached sample. The color can establish a certain relationship to water-holding capacity of the sample. However, it is not easy to judge the changed color of test paper by naked eye. Therefore, color sensor was applied to convert the detected color''s spectrum into specific color parameters, which was manipulated by Arduino controller. Pressure method, as a traditional way to measure water-holding capacity in meat products, is precise but it costs more time and wastes more material. To establish a neural network prediction model of water-holding capacity, this paper set the color parameters as the input vector and set the water-holding capacity measured by the pressure method as the output value. Experimental data derived from 80 samples, which included 60 training samples and 20 testing samples. The back propagation (BP) neural network was trained by 60 samples. Besides, the accuracy of BP neural network was verified by 20 samples. However, the BP neural network had the deficiencies of insufficient network global research ability, slow convergence and local optimum iteration. The genetic algorithm optimized the weights and thresholds in BP neural network, and thus enhanced the accuracy of prediction. The fitness function of genetic algorithm was the sum of square error. After 100 iterations, the best fitness function was obtained. The weights and thresholds optimized by the genetic algorithm were put into the BP neural network, the BP neural network was retrained and the accuracy was tested. The result showed that the optimum attachment time of test paper was 20 s. The optimized BP neural network model based on genetic algorithm had a better ability for nonlinear approach. The determination coefficient of the regression line is 0.987, and the slope of the best linear regression equation is 0.96. This showed that the deviation between the predicted value and the actual measurement value of the BP neural network optimized by the genetic algorithm is very small. Thus the optimization of the model is successful. The prediction accuracy of the BP neural network model was improved from 90% to 95% after being optimized by genetic algorithm. Compared with the pressure method, using color sensor not only greatly shortened the detection time, but also reduced the waste of resources in the detection process. What was more, the cost of color sensor method was lower than NIR (near infrared) spectroscopy method. The predicted results from the optimized BP neural network based on genetic algorithm were better than BP neural network. This detection method is fast and accurate and has low cost. The results provide a reference for further development of intelligent detection equipment for meat products.