基于神经网络的籽棉颜色分级检测

为解决籽棉颜色分级问题,构造了一个基于L^*a^*b^*颜色空间的色度检测仪,主要由颜色传感器、光源及外围电路构成。针对用于籽棉颜色等级检测2个关键指标(反射率、黄度)输出不稳定问题,采用了4层BP神经网络和5块标准色板进行反复训练,使得校准后的反射率的变异系数小于0.21%,黄度的变异系数小于1.13%。在籽棉颜色等级检测实验中,制作了覆盖12个颜色等级的480个测试样。经过反复实验发现,使用该色度检测仪对1个测试样品,需要均匀分布10个测量点结果的平均值,才能得到稳定的色度测量值。最后,采用神经网络方法,对480个籽棉试样数据进行分析,其中:80%用于训练;20%用于识别。实验结果表明,对12个颜色等级的480个样品进行测试,得到的检测准确率都超过了90%。

Research on color grading of seed cotton based on neural network

In order to solve the problem of color classifying for seed cotton, a detection device based on L ^*a^*b^* color space, which is mainly composed of color sensor, light source and peripheral circuit, was designed. Aiming at the unstable output of reflectance and yellowness, which are very important for color grading, a four layers BP neural network was used and trained repeatedly with 5 standard color boards. After the calibration, the rectified reflectance has coefficient of variation with less than 0.21% and the rectified yellowness has coefficient of variation with less than 1.13%. In the next experiment,480 specimen of seed cotton which cover up 12 color grades is prepared. After repeated experiments, it is found that the average value of 10 measurement points, which are evenly distributed in one specimen, could be used as color value for a test sample. Finally, a neural network was applied to analyze the 480 seed cotton sample data, of which 80% was used for training and 20% for identification. Experiment results show that the detection accuracy is more than 90% for all 12 color grades.