基于机器视觉自然场景下成熟柑橘识别

采用2R-G-B色差分量,通过Ostu自适应阈值算法进行图像分割,利用形态运算消除分割后随机噪声,并对分割区域进行标记,利用区域面积和区域最小外接矩形长宽比参数进一步去除背景区域.对于多果重叠问题,利用T=Sqrt(S×I)形成新的图像提取边界,再结合形态学运算实现分割.最后利用优化的圆形Hough变换提取目标图像的形心坐标及半径,恢复遮挡果形.经验证有95%果实能正确识别.

Recognition of mature oranges in natural scene based on machine vision

Using the image of 2R-G-B, Ostu algorithm was used to segment the images of mature oranges and background. Morphologic operation was used to remove the random noise and the object regions were labeled. According to the areas of the labeled regions and their ratios of the least rectangles' length and width, the remnant backgrounds were eliminated. For the overlapped fruits, a new image generated through formula T=Sqrt(S×I) and the boundary of the new image was extracted, with the additional morphologic operation, the overlapped fruits were separated. Finally, the optimized Circular Hough transform algorithm was used to extract centroid coordinates and radius, and then the fruit shape was recovered. The result shows that the correct recognition rate is up to 95%.