果园环境中成熟杨梅自动检测方法研究

Study on Automatic Detection Method of Mature Bayberry in Orchard Environment

针对果园环境中自动摘果机视觉系统受光照、遮挡等影响,导致成熟杨梅检测准确率低的问题,提出一种基于机器视觉的成熟杨梅自动检测方法。首先,利用RGB摄像头采集果园环境中的杨梅图像,并采用基于YUV颜色空间Y分量的限制对比度自适应直方图均衡化算法对采集图像进行处理,以补偿光照对成像质量的影响;然后,采用“G-R”,“2R-G-B”色差法分别提取背景和前景区域,并通过形态学、区域生长等操作,建立基于OneCut算法的前景、背景区域标记控制点集,从而自动分割成熟杨梅区域;最后,针对重叠杨梅区域采用标记控制的分水岭变换方法与凸壳理论进行果实分离,实现每颗果实区域的准确定位。通过试验验证:该方法能准确检测果园环境中成熟的杨梅果实,平均准确率为93.5%,平均召回率为90.6%,具有一定的实用价值。

In view of the problem that the vision system of the automatic fruit picker in the natural environment of the orchard is affected by the light,occlusion and other factors,which leads to the low detection accuracy of mature bayberry,An automatic detection method of mature bayberry based on machine vision is proposed.First of all,RGB camera is used to collect bayberry image in orchard environment.The input image is processed by contrast limited adaptive histogram equalization(CLAHE)based on Y component of YUV color space to compensate the influence of light on image quality.Then,the"G-R"and"2R-G-B"color difference methods are used to extract the background region and foreground region respectively.Based on the operations of morphology and region growth,the set of control points of foreground and background regions is established,and the region of mature bayberry fruit in natural environment is segmented automatically by OneCut segmentation algorithm.Finally,the watershed transform method and convex hull theory are used to segment the overlapped fruit,so that each arbutus fruit area can be located accurately.The experiment results show that the method can accurately detect the mature bayberry fruit in orchard environment,with an average precision of 93.5%and an average recall rate of 90.6%,which has a certain practical value.