基于多光谱成像和近红外点阵结构光的苹果表面缺陷快速检测方法研究

Detection of apple's surface defects is one of the difficulties in the research of the apple's quality detection based on machine vision. However, the discrimination of stem-end/calyx from the real defects on apple is the most difficult to resolve, and there is still not an effective solution to this problem. In this study, a novel method based on multi-spectral imaging and near-infrared speckle-array encoded structured lighting was proposed to solve the problems mentioned above. The effective wavelengths could be determined by studying the response characteristics of the apple's surface defects in 400-2500nm hyperspectral images. The effective wavelengths that could be used to detect the slight bruise and slight decay are also determined. An intensity transformation method was proposed to correct the non-uniform intensity distribution on fruit's surface due to the fruit surface curvature. A multi-spectral imaging system for the apple's surface defects detection was built. Any speckle's position on the apple's surface could be accurately located through the encoding and coding of the near-infrared speckle-array encoded structured lighting. Then the rapid and accurate identification of the stem-end and calyx regions were realized through the position change of the near-infrared speckle-array on the apple's surface. The research on rapid detection of apple's surface defects using multi-spectral imaging and near-infrared speckle-array encoded structured lighting would lay the theoretical basis and methods for the development of an online detection system.

在基于机器视觉的苹果品质检测研究中，表面缺陷的检测是难点，而区分果梗、花萼与真正的缺陷区域是最难解决且至今仍未有效解决的问题。本项目采用多光谱成像和近红外点阵结构光技术融合的方法实现苹果表面缺陷的检测，并有效区分果梗、花萼和缺陷。通过研究苹果表面缺陷在400-2500nm高光谱图像下的响应特性，确定准确表征苹果表面缺陷的最佳指纹波段图像，并研究轻微损伤和轻微腐烂这两类重要缺陷的识别方法,研究因水果表面曲率变化而造成的图像亮度不均的校正方法，构建基于多光谱成像的苹果表面缺陷检测系统；利用一种新的近红外点阵结构光编解码方法实现苹果上任意光斑的准确定位，通过光斑阵列空间位置的变化信息的快速获取，实现果梗、花萼凹陷区域的快速准确识别；研究多光谱计算机视觉系统融合的苹果表面缺陷快速一体化检测方法。研究成果有望为开发基于多光谱成像和近红外点阵结构光的苹果表面缺陷在线快速检测设备奠定理论依据和方法基础。

在基于机器视觉的苹果品质检测研究中，表面缺陷的检测是难点，而区分果梗、花萼与真正的缺陷区域是最难解决且至今仍未有效解决的问题。本项目采用多光谱成像和近红外点阵结构光技术融合的方法实现苹果表面缺陷的检测，并有效区分果梗、花萼和缺陷。通过研究苹果表面缺陷在400-1000 nm高光谱图像下的响应特性，确定准确表征苹果表面缺陷的最佳指纹波段图像，并研究轻微损伤和轻微腐烂这两类重要缺陷的识别方法。.采用分段主成分分析方法确定了有效检测轻微损伤的3个特征波长780、850和960 nm，采用最低噪声分离变换（MNF）获得了有效检测苹果腐烂的4个特征波长563、611、816和966 nm。开发了基于二色分光片和滤光片的多光谱成像系统对特征波长的有效性进行验证，其中损伤的静态检测精度为91.5%，在线检测速度为3个水果/秒、动态检测精度为74.5%。.研究因水果表面曲率变化而造成的图像亮度不均的校正方法，提出了一种类球形亮度校正方法，利用该方法对R分量图像进行亮度变换，变换后的图像使整个水果表面正常区域灰度趋于一致，而缺陷区域依然保留为低灰度区，增强了缺陷和正常果皮的对比度，有效提高了缺陷的检测精度，检测正确率达到93.8%。.提出了一种基于位置变化的点阵结构光编码方法，并用于苹果果梗／花萼的在线检测。通过分析投射在物体和参考平面上光斑的成像规律，通过分析匹配后的差值矩阵，识别果梗／花萼的位置。在线实验结果表明，在满足实时性要求前提下，果梗/花萼的平均识别正确率可达到93%。.在上述研究的基础上，分别对多光谱成像和近红外点阵结构光检测系统进行改进，开发了新的基于棱镜分光的多光谱成像系统，采用了新的基于ToF（time of flight）技术的深度信息传感器实现深度信息的获取，并提出多光谱成像与深度信息获取技术融合的苹果外部品质一体化检测方法，开发了原型系统。研究成果为开发实用的苹果外部品质在线快速检测设备奠定理论依据和方法基础。