论文部分内容阅读
为了了解高光谱图像中光谱提取区域对果品糖度检测模型精度的影响,本文以“华优”猕猴桃为对象,分别提取了10×10、20×20和30×30(像素×像素)的正方形光谱区域以及样品掩膜图像的平均光谱,对平均光谱进行平滑去噪+标准正态变量变换预处理,用处理后的全光谱建立了预测猕猴桃糖度的偏最小二乘、最小二乘支持向量机、极限学习机和误差反向传播网络模型,分析了光谱提取区域对猕猴桃糖度检测精度的影响规律。结果表明,光谱提取面积的增加能够提升最小二乘支持向量机、极限学习机和误差反向传播网络模型的预测性能。基于猕猴桃掩膜图像的平均光谱所建立的最小二乘支持向量机模型具有最好的预测性能,其预测相关系数为0.97,预测均方根误差为0.86oBrix,相对预测误差为4.06。研究说明在高光谱图像中选择合适的光谱提取区域有助于提高模型的预测精度。
In order to understand the influence of the spectral extraction region on the precision of the confectionery detection model in hyperspectral imagery, we extracted 10 × 10,20 × 20 and 30 × 30 (pixel × pixel) Square spectral region and the average spectrum of the sample mask image, smoothing denoising + standard normal variable transformation preprocessing of the average spectrum, the partial least squares, the least square support vector Machine, extreme learning machine and error back propagation network model, the influence of spectral extraction area on the detection accuracy of Kiwi sugar content was analyzed. The results show that the increase of spectral extraction area can improve the prediction performance of least squares support vector machine, extreme learning machine and error back propagation network model. The least square support vector machine model based on the average spectrum of Kiwi mask image has the best prediction performance with a prediction correlation coefficient of 0.97, a prediction root mean square error of 0.86oBrix and a relative prediction error of 4.06. The research shows that selecting the appropriate spectral extraction region in hyperspectral images can improve the prediction accuracy of the model.