高光谱成像技术快速识别苹果和橙子的黑斑及损伤区域

Rapid identification of dark spots and damage areas in apples and oranges by hyperspectral imaging technology

中文关键词:  高光谱成像  黑斑区域  损伤区域  归一化光谱指数NDSI  最小噪声分离变换MNF Rotation

英文关键词:hyperspectral imaging technology  dark spot area  damage area  normalized difference spectral index  minimum noise fraction rotation

中文摘要:

      目的 快速准确地识别出苹果和橙子两类水果的黑斑区域与损伤区域。方法 采用高光谱成像技术快速检测两种水果的黑斑区域与损伤区域: 通过对两类水果在550~780 nm波长范围的光谱进行分析; 运用最小噪声分离方法(minimum noise fraction rotation, MNF)分析了苹果和橙子的7个特征值灰度图, 减少了数据处理的计算需求以区别; 运用归一化光谱指数(normalized difference spectral index, NDSI)方法, 构建新型两波段光谱指数, 以便更好利用波长包含的信息。结果 利用MNF Rotation方法发现第4特征值灰度图能较好地识别出两种水果的黑斑区域与损伤区域, 而NDSI方法中构建NDSI (723.6, 673.6)和NDSI (706, 590)两个波长, 再利用灰度密度分割, 利用2个波段进行四则运算实现了水果损伤和斑点的快速识别。结论 以两种方法均可有效地识别水果损伤与斑点区域, 但MNF Rotation方较为复杂, 运算速度较慢, 不适合在工业生产上进行应用, 而NDSI算法具有运算效率高、实现简便的优势, 更适合用于水果质量检测的快速在线识别。

英文摘要:

      Objective To quickly and accurately identify the black spot areas and damage areas of 2 types of fruits, apples and oranges. Methods Hyperspectral imaging technology was employed for the rapid detection of black spots and damage areas on these 2 types of fruits: By analyzing the spectra of apples and oranges in the range of 550–780 nm. The minimum noise fraction rotation (MNF) was applied to analyze the grayscale images of the first 7 eigenvalues for both fruits, which reduced computational demands for data processing to facilitate differentiation. The normalized difference spectral index (NDSI) method was utilized to construct novel two-band spectral indices, thereby making better use of the information contained within the wavelengths. Results Using the MNF Rotation method, it was found that the grayscale image of the fourth eigenvalue could effectively identify the dark spots and damage areas of the 2 types of fruits. In the NDSI method, two wavelength combinations, NDSI (723.6, 673.6) and NDSI (706, 590), were constructed. Then, through grayscale density slicing and arithmetic operations using the two bands, rapid identification of fruit damage and spots was achieved. Conclusion Both methods can effectively identify the damaged and black spot areas of fruits. However, the MNF Rotation method is relatively complex with a slow computing speed, making it unsuitable for industrial applications. In contrast, the NDSI algorithm features high computational efficiency and simple implementation, which renders it more applicable for the rapid online identification in fruit quality inspection.