王 凡 1,2,孟翔宇 1,陈龙跃 2,段丹丹 1,2,3,钱英军 4.基于高光谱成像的苹果损伤检测方法研究[J].广东农业科学,2023,50(7):57-63 |
查看全文
HTML
基于高光谱成像的苹果损伤检测方法研究 |
Research on Apple Damage Detection Based on Hyperspectral Imaging |
|
DOI:10.16768/j.issn.1004-874X.2023.07.006 |
中文关键词: 苹果 损伤 高光谱 光谱指数 迭代自组织数据分析 |
英文关键词: apple damage hyperspectral spectral index iterative self-organizing data analysis |
基金项目:岭南现代农业科学与技术广东省实验室河源分中心自主科研项目(DT20220009,DT20220007,DT20220011);北京市农林科学院财政项目(CZZJ202203) |
|
摘要点击次数: 837 |
全文下载次数: 437 |
中文摘要: |
【目的】研究苹果损伤高光谱特征,建立基于高光谱成像的苹果损伤区域最佳分类模型,为实时、快速、准确地识别苹果损伤提供重要依据。【方法】以北京平谷区收集的苹果样品为研究对象,利用高光谱图像技术检测水果表面机械损伤。利用 390 ~1 000 nm 范围的高光谱图像(HSI)数据,通过比值光谱分析损伤与正常感兴趣区域(ROI)的光谱响应特性,筛选特征波段,并构建较好地突出损伤区域特征的 3 种类型光谱指数:归一化光谱指数(NDSI)、比值光谱指数(RSI)和差值光谱指数(DSI)。在此基础上,优选提取损伤区域能力较强的光谱指数,利用迭代自组织数据分析(ISODATA)无监督据聚类算法提取苹果损伤区域。【结果】当苹果表面受到损伤时,光谱反射率变化显著。波段优化后发现,528、676 nm 的反射率可以有效识别异常区域。基于选定的特征波段,构建苹果损伤检测的识别光谱指数,包括 NDSI、RSI 和 DSI。光谱指数图像的像素值分析发现,损伤区域特征与正常区域特征在各光谱指数(SI)增强图像中区分明显。两类图像特征的 NDSI 像素平均值相差最大、达到 0.629,表明建立的 NDSI 对损伤区域及正常区域特征具有较强的区分能力。利用无监督分类方法 ISODATA 分类,验证了光谱特征指数在检测苹果损伤方面具有较高的特异性,对苹果损伤的检测正确率
达到 92.50%。【结论】研究结果适用于苹果损伤的实时快速检测,为苹果的精准管理生产提供技术基础与参考。 |
英文摘要: |
【Objective】By studying the hyperspectral characteristics of apple damage and establishing the best classification model of apple damaged area based on hyperspectral imaging, we can provide an important basis for real-time, rapid and accurate identification of apple damage.【Method】This paper investigates the use of hyperspectral image technology to detect mechanical damage on fruit surfaces using apple samples collected from Pinggu District, Beijing as the research object. Using the hyperspectral image (HSI) data in the range of 390-1 000 nm, the spectral response characteristics of the damaged and normal ROI areas was analysed by ratio spectroscopy, and three types of spectral indexes, namely the normalized difference spectral index (NDSI), the ratio spectral index (RSI) and the difference spectral index (DSI) were selected and constructed to better highlight the characteristics of the damaged areas..On the basis of these indexes, the spectral indexes with better ability to extract damage areas were selected, and the unsupervised data clustering algorithm of iterative self-organizing data analysis techniques algorithm (ISODATA) was used to extract apple damage areas.【Result】The spectral reflectance changed significantly when the apple surface was damaged. After waveband optimization, it was found that the reflectance of 528 nm and 676 nm could effectively identify the abnormal areas. Based on the selected feature bands, the recognition spectral indexes for apple damage detection were constructed, including NDSI, RSI and DSI. By the pixel value analysis of the spectral index images, the damaged area features were distinguished from the normal area features in each SI enhanced image obviously. Among them, the difference between the NDSI pixel averages of the two types of image features was the largest, reaching 0.629, which indicated that the established NDSI has a strong ability to distinguish between the damaged area and normal area features. Using the unsupervised classification method ISODATA classification, it was verified that the spectral feature index has high specificity in detecting apple damage, and the accuracy rate of apple damage reached 92.50%.
【Conclusion】The results are applicable to the real-time and rapid detection of apple damage, providing a technical basis and reliable reference for the accurate management of apple production. |
查看/发表评论 下载PDF阅读器 |
|
|
|