基于高光谱成像技术的灵武长枣常见缺陷检测

以灵武长枣为研究对象,利用高光谱成像技术结合主成分分析法(principal component analysis,PCA)和最小噪声分离法(minimum noise fraction,MNF)对长枣缺陷进行快速检测与识别,主要探讨样本背景对缺陷识别的影响。首先,采集虫眼、裂痕、正常枣的高光谱图像,利用PCA法和MNF法分别对其降维去噪,选择虫眼与正常枣的PC1和M1图像、裂痕枣的PC2和M2图像进行缺陷识别,经PCA分析后的缺陷识别率均为100%,MNF处理后的识别率分别为69.2%,56.8%,100%;随后对其高光谱图像进行掩模去背景,再对其降维去噪后检测识别,PCA后的识别率均为100%,MNF后的识别率分别为73.1%,65.9%,100%。结果表明:利用高光谱成像技术结合两种降维去噪法对长枣常见缺陷的识别是可行的,背景干扰对于PCA法的缺陷识别不影响,其识别效果优于MNF法,且去背景后的MNF法缺陷识别率较未去背景的有所提高,为后续长枣缺陷的在线检测提供理论依据。

Detection of common defects in jujube fruit using hyperspectral imaging

A hyperspectral imaging technology combined with the principal component analysis (PCA) and the minimum noise fraction (MNF) methods were developed for the detection of common defects in Lingwu long jujubes, and investigated the influence of the background to recognition of defects. Firstly, the hyperspectral images of jujube samples (insect hole, crack and intact jujubes) were acquired. Secondly, the PCA and MNF methods were used to reduce dimensionality of hyperspectral images and to separate the noise from signals effectively. The PC1 and M1 images of insect hole and intact jujube, PC2 and M2 images of crack jujube were selected to distinguish different type of jujubes. By the PCA method, the classification rates of three kinds of jujubes all were 100%. And by the MNF method, the classification rates of insect hole jujubes, crack jujubes and intact jujubes were 69.2%, 56.8%, 100%, respectively. Then, the masked original hyperspectral images were to remove the effect of background and analyzed by the PCA and MNF method again. The classification rates by the PCA method were all 100%, and the classification rates by the MNF method were 73.1%, 65.9%, 100%, respectively. The results showed that the hyperspectral imaging technology combined with PCA and MNF methods were feasible. The influence of the background by the MNF method to defect recognition was slight and the impact to defect recognition by the PCA method gained the advantage over the MNF method. The recognition rate of the MNF method combined with background mask was better than that of no background mask, and to provide the theory basis for the common defects of online detection in future.