自然场景下不同成熟度柑橘的自适应识别

针对果实在未完全成熟前就需采摘以满足市场销售需求的实际情况，以及采摘机器人作业环境的复杂性，研究了自然场景下不同成熟度柑橘的识别方法。通过R与G分量间比值变换得到具有独特信息的比值图像，并在Munsell HSV颜色空间上，利用共轭激励法融合H和S分量图像，针对其与上述比值图像的弱相关性与强正交性，用其构建掩膜，来去除比值图像的背景，实现柑橘识别。实验结果表明此方法对柑橘识别精度达到92％，可以自适应地识别自然场景下不同成熟度的柑橘，且识别精度高，鲁棒性强。

Adaptive Recognition of Different Maturity Citrus in Natural Scenes

In order to meet the fruit demand of market and guide the robots to harvest citrus fruit, a method based on image processing is developed to identify different maturity citrus fruits in complex natural scenes. Ratio transformation between the R and G component is attempted to acquire a ratio image. In the Munseil HSV color space, conjugate incentive algorithm is attempted to fuse H and S component to acquire a fusion image, which is weakly correlated with the ratio image. A mask is built by the fusion image to remove the non-citrus background of the ratio image. Experimental results show that citrus can be recognized with an accuracy of 92%. This work demonstrates that the adopted method in this paper is very robust and effective to adaptively recognize different maturity citrus in natural scenes.