【作者】 蔡健荣； 赵杰文；

【Author】 Cai Jian-rong, Zhao Jie-wen (School of Biological and Environmental Engineering, Jiangsu University, Zhenjiang 212013, China )

【机构】 江苏大学食品与生物工程学院；

【摘要】 利用圆形Hough变换,分割和恢复相互重叠、部分遮挡的水果果形,并提取果实位置和大小特征。针对Hough变换耗时长的问题,在进行Hough变换时,采用分割得到的目标区域面积,自适应调整搜索半径的区间和累加器的单元尺寸,解决了因目标区域面积增加而计算量增加和累加器容量增加的问题。采用高斯算子对变换后的累加器数据进行平滑处理,以提高信噪比;设置邻域系数以避免同一水果区域内出现多个目标;设置停止搜索阈值以避免背景区域出现虚假目标;最后结合目标区域的位置信息以辨别检测得到目标的真伪性。检测程序计算得到的目标参数,能准确反映果实的位置和大小。其检测准确率为84．8％, 重叠或遮挡面积小于1／3的水果能准确识别。更多还原

【Abstract】 The circular Hough transformation (CHT) is used to recover the shape of overlapped fruits and extract the location and size features of fruits. As the CHT algorithm takes a mass of CPU time, the area of target region was used to adjust the cell size of accumulator to adapt the image size. This method can restrain the increment of compute time and accumulator’s capacity along with increasing of the area of the target region. During data processing, Gaussian filter was firstly used to smooth the accumulator’s data to improve Signal-to-Noise; secondly, the neighbor coefficient was set to avoid multi-targets inside same fruit region, thirdly terminating threshold was set to prevent false object appearance in background; finally, target’s location and area information was combined to identify true objects from candidates. By our experiment, the validity reaches 84.8%, and the fruits can be correctly detected when overlapped area is less than 1/3 of entire area.更多还原