一种融合动态OTSU和几何特征的苹果视觉分割算法

A visual segmentation algorithm for apples based on dynamic OTSU and geometrical features

为了实现苹果自动采摘过程中视觉图像的准确分割,提出了一种融合动态OTSU和几何特征的苹果视觉分割算法。该算法首先将苹果图像从RGB颜色空间转换到Y、Cb、Cr分量加权的颜色空间;然后进行动态OTSU阈值分割;最后对分割出的二值图像建立连通区域,并利用目标区域与干扰区域的面积和类圆度的几何特征差异完成苹果目标的识别。四种不同环境下的实验测试结果表明,与其他相关算法相比,融合动态OTSU和几何特征的视觉分割算法不仅保证了高检出率,而且能有效地将误检率平均降低至21.8%、噪声残余率平均降低至20.0%,从而验证了文中算法的有效性。

In order to achieve accurate visual segmentation during the automatic apple picking process,a visual segmentation algorithm for apples based on dynamic OTSU and geometrical features is proposed.First,an apple original image was transformed from a RGB color space to a color space with the weighted Y,Cb and Cr components. Then a dynamic OTSU threshold segmentation was performed. Finally,the connected regions were established in the segmented image and the apple targets were distinguished using the differences of area size and circularity between target regions and interference regions. The experimental results in four different external environments show that, compared with other related segmentation algorithms,the visual segmentation algorithm for apples based on dynamic OTSU and geometrical features can not only guarantee a high detection rate,but also reduces its false detection rate to 21. 8% and decrease the noise residual rate to 20. 0% effectively,which verifies the validity of the proposed algorithm.