基于CARS与PCA的高光谱煤岩特征信息检测方法

Coal and rock feature detection method based on CARS and PCA

针对采煤机智能化截割时存在煤岩识别精度低、稳定性差等问题,提出一种基于高光谱成像技术的煤岩检测方案。使用8种不同类型的煤岩样本(训练集800块、预测集200块)进行分析,利用竞争性自适应重加权算法将光谱全波段降维至11个特征波长形成光谱特征向量;通过灰度共生矩阵来描述煤岩的纹理特征,选取对比度、能量、同质性3个特征参数值作为纹理特征向量;通过主成分分析融合剔除光谱与纹理特征中解释能力较差的特征信息,利用预测集样本分别对光谱全波段、CARS提取特征波长、图像纹理、CARS提取特征波长融合纹理特征、光谱全波段融合纹理特征和PCA融合特征波长与特征纹理特征的特征向量建立偏最小二乘回归模型,通过对比6种特征向量的建模预测性能,选出煤岩最优特征向量。PCA算法融合后特征向量预测性能的R2,RMSE,平均绝对误差MAE和准确率分别为0.912,0.201,0.151和94%.该方法可改善煤岩特征信息检测的稳定性与可靠性,为煤岩识别提供有效的特征信息,对实现采煤机智能化开采具有重要意义。

Due to the low accuracy and poor stability of coal and rock identification in intelligent cutting of shearer,a coal and rock detection scheme based on hyperspectral imaging technology was proposed.Eight different types of coal and rock samples were chosen with 800 training sets and 200 prediction sets for experimental analysis.Firstly,Competitive Adaptive Reweighted Sampling(CARS)algorithm was used to reduce the spectral full-band to 11 characteristic wavelengths to form a spectral eigenvectors.Secondly,the texture features of coal and rock were described by the second order statistical gray level co-occurrence matrix,and three feature parameters of contrast,energy and homogeneity were selected as texture feature vectors.Finally,the feature information with poor interpretation ability in spectral and texture features was fused and eliminated by Principal Component Analysis(PCA).The partial least square regression model was established by using the prediction set samples to extract the full spectral band,the CARS extraction feature wavelength,the image texture,the CARS extraction feature wavelength fusion texture feature,the full spectrum band fusion texture feature,the PCA fusion feature wavelength and the feature vector of the texture feature.By comparing the modeling and prediction performance of six feature vectors,the optimal feature vector of coal and rock was selected.The results show that the R2,RMSE,average absolute error MAE and accuracy of feature vector prediction performance after fusion of PCA algorithm are 0.912,0.201,0.151 and 94%,respectively.The stability and reliability of coal and rock characteristic information detection could be improved by this method,provideing an effective feature information for coal and rock identification,which is of great significance to realize intelligent mining of shearer.