水果直径和缺陷面积的机器视觉检测

摘要

Huanghua pear is an important fruit in China. The fruit diameter and surface defects are important indices in the classification of Huanghua pear. To detect the surface quality of Huanghua pear, a machine vision system was set up on the basis of investigating the spectral reflectance of the Huanghua pear. The effects of different backgrounds on the acquired images were studied, and the method to seek the optimum background in the light of the grayscale histogram was found to be very effective. Clearer images were acquired when the background was white. As we know, fruits in actual situation are commonly random in orientation. Therefore, the Minimum Enclosing Rectangle (MER) method was designed and used to estimate the maximum diameter, and the correlation coefficient of real maximum diameter versus the maximum diameter measured by MER method reached 0.9962. According to color difference in the neighboring area of the defected and non-defects, the grayscale values of the R (Red) frame and G (Green) frame were used to find the suspectable defects, and the whole defect was found by the region-growing method. To decrease the relative errors, the pixel transform method was adopted to recover the geometrical feature of sphere-fruit surface from projected image while the area of defect was calculated. Moreover, a method to revise the estimated area was advanced. Compared to the method to directly calculate the real area of defect from the projected image, the pixel transform method could decrease the relative error by about 38%. These results lay a solid foundation for further development of a Huanghua pear quality detection system with machine vision.%黄花梨是中国的一种重要水果,果径和果面缺陷面积是黄花梨分级的两项关键指标.通过研究黄花梨的分光反射特性,研制了一套适合黄花梨品质检测的机器视觉系统.为了适应实际生产中水果方向的随机性和水果外形的不规则性的要求,使水果尺寸检测的方法有更好的适应性,设计了一种利用水果的最小外接矩形(MER)法求最大横径的方法,并进行了试验验证,得出了表示实际最大横径与预测最大横径的关系的回归方程式,两者的相关系数为0.9962.分析了黄花梨缺陷区域的R、G、B各分量灰度的变化特点,利用R分量灰度和G分量灰度在缺陷区域和完好区域交界处有明显突变这一特点,采用梯度算法求得了可疑缺陷点,然后再用区域生长法,找出了缺陷点像素的最大连通集及所有的缺陷区域;采用像素点变换法,实现了根据三维物体的二维投影图像恢复物体表面的真实几何面积的设想,大大降低了缺陷面积计算的误差;另外,还提出了一种新的面积修正方法,即用实际缺陷面积等于经像素点变换后的缺陷面积减去缺陷区域周长的一半加上1个像素点的面积来进行修正,进一步提高了缺陷面积计算的精度,而且该修正方法同样适用于其它图像面积的计算.