基于颜色与面积特征的方格蔟蚕茧分割定位算法与试验

Algorithm and Experiment of Cocoon Segmentation and Location Based on Color and Area Feature

蚕虫上蔟多采用纸板方格蔟,但纸板方格蔟在使用过程中会因扭曲变形导致方格分布不规则,而采茧机械对变形的方格蔟进行蚕茧采摘时,会对方格蔟造成损伤。为了提高方格蔟机械采茧的智能化水平,减少采茧设备对方格蔟的损伤,提出一种基于颜色与面积特征的方格蔟蚕茧分割定位算法,实现对方格蔟中蚕茧的分割、中心点定位和位置坐标的视觉测量。首先采用图像空间的Brown畸变模型对方格蔟图像进行畸变矫正,减小径向畸变对视觉测量的影响;对矫正后的图像采用Mean Shift聚类算法进行预分割,消除光照及图像背景对蚕茧分割的影响;然后对阈值分割和形态学处理后的二值化蚕茧图像进行基于面积特征的连通域标定,得到每个蚕茧中心点位置;将连通域标定得到的蚕茧中心点坐标代入图像坐标系与世界坐标系转换方程,得到每个蚕茧在笛卡尔空间的三维坐标,经过视觉测量确定蚕茧在方格蔟中的具体位置,控制蚕茧采摘装置采摘方格蔟中的蚕茧。经过试验,该算法对方格蔟中的蚕茧检测正确率为96.88%,蚕茧坐标最大定位偏差小于6.0 mm,满足采茧装置对蚕茧采摘的定位精度要求。

To improve the lower efficiency of silkworm cocoon harvesting,an algorithm of cocoon image segmentation and coordinate location was proposed based on color and area characteristics,and a cocoon harvestor was designed based on machine vision. The monocular CMOS camera was firstly used in the algorithm to take image of checker cocooning frame. And the non-measurement distortion correction method was used to correct the image. Secondly,the camera model was calibrated with the internal parameters for the monocular two-dimensional visual measurement system. The image was smoothed via gray and mean shift filter method because the outer floss of the cocoon can cause wrong segmentation of the image in checker cocooning frame image. Then the binary image was obtained by threshold segmentation. Next,the binary image was processed by open operation and area feature extraction method to remove noise region. A part of the smaller noise connected components can be removed by the open operation. The cocoon region can be extracted by the area characteristic when the large area of the connected components can be removed. The center point coordinates of the cocoon region were got by the connected components calibration,and were mapped into the world coordinates through the equation that transformed image coordinates to world coordinates to get the cocoons＇ positions in the Cartesian space.Finally,the cocoons were harvested by the cocoon harvestor. According to the experiment,the algorithm had the accuracy rate of 96. 88% for the cocoon detection in the checker cocooning frame and less than 6. 0 mm for the cocoon coordinate,which satisfied the requirement of the location of cocoon harvesting.