基于超红图像与轮廓曲率的苹果目标识别与定位方法研究

Recognition and Localization of Apples Based on Super-red Image and Contour Curvature

为实现苹果目标的识别及其空间定位,提出了一种自然场景下苹果目标的识别与定位方法。该方法首先将RGB颜色空间转换至HIS颜色空间以得到自然场景下苹果图像的色调分量H和饱和度分量S,为了充分利用其色调信息,采用了基于超红图像的苹果目标识别方法并应用基于区域的分割方法实现了目标的有效分割;接着利用轮廓曲率法抽取连续光滑的轮廓曲线并估计该光滑曲线段的圆心及其半径参数,实现果实的定位;最后利用逐行扫描法,结合苹果生理特性,实现了苹果采摘点的有效定位。为了验证算法的有效性,利用50幅富士苹果图像进行了试验。试验结果表明,苹果及其果柄的识别率在80%以上,对于轻度遮挡的苹果目标,基本满足其定位要求。

Aiming at realizing the recognition and localization of apples, a suitable algorithm that could be used in nature scene was proposed. On the first stage, the image was transformed from RGB color space to HIS color space, and the hue component H and the saturation component were got. To make full use of the hue information, a super-red theory was applied to segment apples. And then, the apple regions were got by combining region-based segmentation and mathematical morphology theory. After that, smooth contours were extracted by using contour curvature method and the center and radius parameters were estimated, which were essential for the localization of apples. On the final stage, apple picking points localization were realized by using progressive scanning method combined with apples’ physiological characteristics. To validate the algorithm, the experiments were carried out by using 50 Fuji apples. The results illustrated that the recognition rate of the apple and its stalk was above 80%. And the method proved to be feasi-ble and effective for slightly occluded apples.