苹果糖度高光谱图像可视化预测的光强度校正方法

Intensity Correction of Visualized Prediction for Sugar Content in Apple Using Hyperspectral Imaging

针对类球形水果表面曲率变化引起高光谱图像光响应强度差异较大,难以有效预测各部位的品质信息的问题,以富士苹果为研究对象,对高光谱图像进行黑白标定后,以糖度测试部位为感兴趣区域提取平均光谱并建立糖度的定量预测模型,校正集相关系数Rc为0.930 5,校正均方根误差RMSEC为0.433 1;高光谱图像经构建掩模消除样本背景噪声后,提出了高光谱图像光强度校正方法,比较校正前后的高光谱图像能量分布图可以发现光强度得到有效补偿,对校正后的高光谱图像标记空间信息并提取对应光谱,用已建立的苹果糖度模型计算各像素点对应的糖度值,绘制苹果糖度的伪彩色分布图。研究结果表明,高光谱图像经强度校正可以快速无损的预测苹果的糖度及其分布。

Hyperspectral imaging which integrating both spectroscopic and imaging techniques with higher spatial and spectral resolution,has been developed to study the physical characteristics, chemical constituents and distributions of different quality attributes. It＇s difficult to further analyze because of the adverse effects produced by the curvature of spherical objects in the process of hyperspectral images acquirement. Its suitability was illustrated in a specific case of apple fruits. This study proposes a method for correcting the light intensity of radiation non-uniform on the apple fruits. Firstly,the original hyperspectral images were corrected into the reflectance hyperspectral images based on black and white reference images,resulting in reducing the influence of illumination and the dark current of the camera.Then,the mean spectra extracted from roundness region of interest（ ROI） in centre area of hyperspectral image were used to develop calibration models by using partial least squares（ PLS） regression. The correlation coefficient and root mean square errors of calibration were found to be 0. 930 5 and 0. 433 1,respectively. After applying the proposed correction,the spectra of the pixels in hyperspectral image were performed to calculate the sugar content of corresponding pixels. Finally,the visualization of sugar content distribution in apple was achieved by using pseudo-color mapping. The results demonstrated that the correction method was proved to be effective for eliminating the adverse effects produced by thecurvature of the fruit on the intensity of the radiation. The hyperspectral imaging has a great potential to be a nondestructive and rapid tool for the quantitative measurement of sugar content distribution for apple.