论文部分内容阅读
目的利用高光谱技术检测苹果外观缺陷,分析主成分分析法和波段比率算法研究高光谱图像的可行性。方法在400～1100nm波长范围内获取苹果表面的高光谱图像信息,用主成分分析法处理高光谱下采集的苹果图像,选取第三主成分图像进行分析,作为最后的判别依据。波段比率算法中选取了717nm和530nm两个有效波段,将两个波段的图像进行比值运算。717nm波段的图像进行阈值运算、中值滤波及形态学分析得到二值化掩膜图像,再与二值化后的比率图像进行布尔运算,提取缺陷的有效信息。结果基于主成分分析法,检测苹果表面缺陷的分级准确率为81.25%,波段比率算法对苹果表面缺陷的分级准确率为93.75%。结论利用高光谱成像技术下波段比率算法相对于主成分分成法更适合于实时、在线、快速检测。
Objective To detect the appearance defect of apple by using hyperspectral technique and to analyze the feasibility of studying hyperspectral image by principal component analysis and band ratio algorithm. Methods The hyperspectral image information of apple surface was acquired in the wavelength range of 400-1100 nm. The principal component analysis was used to process the images of apple collected under hyperspectral, and the third principal component image was selected for analysis. Band ratio algorithm selected two effective bands 717nm and 530nm, the ratio of the two bands of image computing. 717nm band images were thresholding, median filtering and morphological analysis to obtain a binary mask image, and then with the ratio of the binary image Boolean operation to extract the effective information of defects. Results Based on the principal component analysis, the grading accuracy of detecting apple surface defects was 81.25%. The accuracy of the band ratio algorithm was 93.75%. Conclusion The algorithm of band ratio under hyperspectral imaging is more suitable for real-time, online and rapid detection than the principal component analysis.