论文部分内容阅读
以新疆红富士苹果为研究对象,初步探讨应用高光谱图像技术预测其重量的研究方法。首先采用高光谱成像仪采集苹果侧面和赤道面投影图像,提取苹果高光谱图像中前景与背景光谱偏差最大的波长图像(733 nm),对其图像分割后获得目标区域的面积(像素数)特征。随后,采用不同的特征和建模方法,建立不同的重量预测模型,对比后确定最佳模型。结果表明,采用两个体积特征建立多元线性回归重量预测模型,苹果重量预测值与实际值间相关系数为0.9927,预测均方根误差为4.3393 g。
Taking Fuji apple in Xinjiang as the research object, the research method of using hyperspectral imaging technology to predict its weight was discussed. First, the images of apple lateral and equatorial planes were collected by hyperspectral imager. The wavelength (733 nm) with the largest deviation between the foreground and background spectra was extracted and the area (number of pixels) of the target area was obtained after image segmentation . Subsequently, using different features and modeling methods, different weight prediction models were established, and the best model was compared. The results showed that the multivariate linear regression weight prediction model was established by using two volumetric features. The correlation coefficient between predicted apple weight and actual value was 0.9927, and the root mean square error of prediction was 4.3393 g.