【作者】 周水琴；

【导师】 应义斌；

【作者基本信息】 浙江大学 ， 农业机械化工程， 2004， 硕士

【摘要】 在机器视觉技术对水果表面颜色检测和自动分级过程中，水果图像质量直接影响最终的检测与分级结果。由于机器视觉系统硬件设备的局限性及光照环境引起的图像色度失真不可避免，应采用适当的校正模型予以校正。本文主要针对图像的色度失真，研究了机器视觉系统的色度失真规律，建立了色度失真校正模型，并应用于水果颜色分级，验证了色度失真校正模型的有效性。主要研究内容和研究结果如下： 1、综述了利用机器视觉技术进行水果品质无损检测和自动分级的国内外研究进展，并指出了国内外同类研究中所存在的问题。 2、建立并完善了适合本研究的机器视觉系统。该系统由MeterⅡ／MC（Matrox．Inc．）图像采集卡、TMC-7DSP（PULNIX）CCD及6支F40BX／840（GE）荧光灯组成。 3、以平面色卡为研究对象，通过比较机器视觉系统与色差计测量所得色度值，分析了机器视觉系统的图像色度失真规律，并建立了色度失真校正模型。 4、对27个平面色卡在3个不同位置作对比研究，建立了平面色卡色度失真校正模型：0位置：y₀=1.0318x₀-3.3828，模型相关系数R²=0.9817；1位置：y₁=1.0336x₁-2.8485，模型相关系数R²=0.9859；2位置：y₂=1.0304x₂-2.2285，模型相关系数R²=0.9890。其中，y₀、y₁、y₂和x₀、x₁、x₂分别是三个位置校正后与校正前的色度。 5、分别在上述三个位置采集10种颜色的球面图像，验证了在本研究所用机器视觉系统内，物体表面曲率对色度失真基本没有影响，因此，对于球面物体的色度失真，可直接应用相应位置的平面色卡色度失真校正模型进行校正。此研究结论对于其它机器视觉系统的色度校正有一定的借鉴意义。 6、以HSV颜色模型中色度值为指标，研究了水果表面的色度分布范围在35°-80°之间，把色度按10°划分成五个区域。分别以[35°，45°]、[34°，44°]和[36°，46°]区间内色度累计频度作为着色依据进行颜色分级，发现[36°，46°]色度区间计算机分级与人工分级的结果一致度最好。 7、用与水果相应位置的色度失真校正模型对水果表面色度进行校正后，在[36°，46°]区间内，校正后水果一、二、三等及等外品颜色分级的正确率分别为100％、75.00％、75.00％、100％，而校正前分别为71.43％、73.33％、46.67％和100％。从而验证了机器视觉系统色度失真校正模型和方法的有效性。更多还原

【Abstract】 In the process of color inspection and automatic grading of fruits with machine vision, the quality of fruit image captured by cameras would influences the accuracy of inspection and grading directly. It is impossible to completely avoid the color distortion caused by the hardware of machine vision system and the illumination of lighting chamber, so if should be corrected with proper correction model. The laws of color distortion of the machine vision system were studied, the color distortion and correction model is set up, and the validity of model in fruit color grading is proved. Main contents and results were as follows:The research advancements and achievements in the area of nondestructive inspection of fruit quality and automatic grading with machine vision were reviewed, and the existing problems were put forward.Machine vision system suitable for this research was Set up . This system is composed of Meter II / MC ( Matrox.Inc. ) image grabber, TMC-7DSP (PULNIX) CCD and 6 F40BX/840 (GE) fluorescent lamps.Hue of flat cards got form machine vision system and color meter was compared. Law of hue distortion with machine vision system was analysed and hue correction model was set up.hue of 27 flat card with different colors captured at 3 different positions was compared and hue correction model was set up. Position 0: y0 =1.0318x0-3.3828 , position 1: y1 =1.0336x1 -2.8485 , positions 2: y2 =1.0304x2 -2.2285 . Model correlation coefficient R2 was 0.9817,0.9859 and 0.9890 respectively. Here, y0 , y1 , y2 and x0 , x1 , x2 were the hue value after and before correction at three position respectively.The sphere images of 10 balls were captured at 3 different positions as the flat cards. It is proved that curvature of object have no influence on hue distortion in this machine vision system, basically. So distorted hue of balls can be corrected by the model of corresponding position flat cards directly. This conclusion can be referenced by other machine vision system.Regarding hue value in HSV color model as index, hue of fruit surface ranges from 35° to 80°, which was divided into five areas with 10° intervals. Fruits were graded according to hue frequencies of [35°, 45°] [34°, 44°] and [36°, 46°] respectively .The best Result was got at area [36°, 46°].After correcting with hue distortion model at the corresponding position, the grading accuracies of grade 1, grade 2, grade 3 and substandard fruit were 100%, 75.00%, 75.00% and 100% respectively while 71.43%, 73.33%,46.67% and 100% respectively before correction.It was proved that the hue correction model and way with machine vision system were effective.更多还原