不同生长状态下多目标番茄图像的自动分割方法

Automatic segmentation method for multi-tomato images under various growth conditions

将自然生长状态下的成熟果实从复杂背景中识别出来并确定其空间位置,是实现果实采摘作业智能化的基础。该文针对在自然光照条件下多个番茄自然生长状态为相互分离、靠拢或重叠以及被枝叶部分遮挡的情况,研究了一种成熟番茄图像的自动分割方法。该方法利用RGB颜色空间下番茄图像中目标与背景的(R-G)灰度值存在明显差异的特点,首先使用O tsu法对番茄的RGB彩色图像的色差灰度图像(R-G)进行动态阈值分割,然后对番茄的R分量灰度图像应用基于形态重建的受控标记分水岭算法搜索靠拢或重叠番茄的分界线,最后对前面两次运算的结果作交集运算得到最终分割的二值图像,将番茄从背景中分割出来。通过100幅番茄图像进行试验表明,该方法不仅能对自然光照条件下不同生长状态的多目标番茄图像进行有效分割,而且对番茄的成熟度及品种差异也具有很好的鲁棒性。

It is fundamental to realize intelligent fruit-picking that mature trusts are dxstxnguxshed trom complicated backgrounds and their three-dimensional locations are determined. Various methods for fruit identification can be found from the literature. However, surprisingly little attention has been paid to image segmentation of multifruits which growth morphologies are connected, overlapped and partially covered by branches and leaves of plant under natural illumination conditions. In this paper the authors present an automatic segmentation method that comprises three main steps. First, Red and Green component images are extracted from RGB color image, and Green component subtracted from Red component gives RG of color aberration gray-level image. Gray-level value between objects and background has obvious difference in RG image. By the feature, Ostu＇s threshold method is applied to do adaptive RG image segmentation. And then, controlled-watershed segmentation based on morphological grayscale reconstruction is applied into Red component image to search edge boundary of connected or overlapped tomatoes. Finally, intersection operation is done by operation results of above two steps to obtain black and white image of final segmentation. The tests show that the automatic segmentation method has satisfactory effects upon multi-tomato images of various growth morphologies under natural illumination conditions. Meanwhile, it is very robust for different maturities of multi-tomato images.