机器视觉在生球团含水率预测中的应用

Application of machine vision in water content rate prediction of green pellets

为了实现对生球团含水率的无接触式快速检测,建立生球团含水率预测模型。以铁精矿生球团为研究对象,利用中值滤波器去除图像噪声,再提取生球团图像的灰度直方图特征(最大概率灰度、平均灰度、标准方差、平滑度、标准偏差、峰态、偏斜度)及灰度共生矩阵纹理特征(能量、熵、对比度、相关性),分别以其为输入指标,建立粒子群优化的支持向量机回归预测模型对含水率进行预测,比较不同输入特征的预测精度。结果表明:灰度直方图特征预测结果的平均绝对误差和平均相对误差分别为0.037 4和0.524,灰度共生矩阵纹理特征预测结果的平均绝对误差和平均相对误差分别为0.020 1和0.284 5;灰度共生矩阵纹理特征预测精度高于灰度直方图特征预测精度。

A water content rate prediction model of the green pellet is established to realize non?contact rapid detection for the water content rate of green pellets.Taking the green pellets of iron ore concentrate as the research object,the median filters are used to remove image noises.The gray histogram features(the maximum probability gray value,average gray value,stan-dard variance,smoothness,standard deviation,kurtosis,and skewness)and the gray-level cooccurrence matrix(GLCM)tex-tural features(energy,entropy,contrast,correlation)of green pellet images are extracted.Taking the extracted features as input indexes,the support vector machine regression prediction model based on particle swarm optimization is established to predict the water content rate and compare the prediction precisions of different input features.The results show that the average absolute error and average relative error for prediction results of gray histogram features are 0.037 4 and 0.524 respectively,while the average absolute error and average relative error for prediction results of gray-level cooccurrence matrix textural features are 0.020 1 and 0.284 5 respectively,which indicates that the prediction precision of graylevel co?occurrence matrix textural fea?tures is higher than that of gray histogram features.