基于改进RHT及均值漂移聚类方法的双孢菇图像目标提取研究

Research on target extraction of Agaricus bisporus image based on improved RHT and Mean Shift clustering method

工厂化生产环境下双孢菇图像存在成簇粘连、菌柄倾斜、白色菌丝繁盛等复杂状态,采用常规阈值分割方法无法准确提取目标,采用改进的随机霍夫变换圆形检测算法与滑动平均聚类算法结合的方法开展目标提取研究。通过图像预处理降低部分背景干扰,改进边缘点取样方式提升圆形检测效率,依据菌盖半径范围剔除无效参数;以双孢菇图像检测出的所有圆形为目标,采用均值漂移聚类算法合并圆形,消除圆形检测算法中识别出的错误目标。采用算法提取目标结果:针对双孢菇规则生长、菌盖附土、菌盖重叠率不高或变形较小的图像,目标提取正确率大于92%;针对大面积菌丝包围菌菇、菌菇成簇倾斜、大小菌菇密集掺杂、菌盖堆积变形严重的图像,存在圆形检测不准、聚类误判或漏判等问题,目标提取正确率大于82%。本文采用的算法数据处理量小、计算速度快、适应性强,能够满足菌菇生产过程中长势自动监测、出菇统计等需求。

In the factory production environment,Agaricus bisporus images have complex states such as cluster adhesion,tilted fungiculum and abundant white mycelia,which cannot be accurately extracted by conventional threshold segmentation method.Under such background,this paper combined an improved Random Hough Transform(RHT)and randomized circle detection algorithm(RCD)with the moving average clustering algorithm(MVCA)for target extraction.Some background interference was reduced through image preprocessing.The sampling method at the edge points was improved to increase the circle detection efficiency.The invalid parameters were eliminated based on the radius of the pileus.All the circles detected by the Agaricus bisporus images were taken as the target.Then MVCA was used to merge the same circles and eliminate erroneous targets identified in the circle detection algorithm.The algorithm was used to extract the target results.The accuracy of target extraction was over 92%for images of Agaricus bisporus with regular growth,pileus covered by soil,low overlapping ratio or small deformation of pileus.While for images of Agaricus bisporus surrounded by large-area hyphae,tilting in clusters,uneven in size or with serious pileus piling up and deformation,the target extraction accuracy was more than 82%,and there were problems of inaccurate circle detection,misjudgment of clustering or missed judgment.With the advantages of small amount of algorithm data,fast calculation and strong adaptability,the RHT-RCD-MVCA integrated method proposed in this paper could meet the needs of automatic monitoring and fruiting statistics during the growth of the Agaricus bisporus.