基于叶片匹配的作物植株图像分离与计数方法

Individual plant extraction and counting from field images based on leaf matching

构建了从连续采集的田间图像中提取作物植株的算法,为图像采集植株微观生长信息提供算法支撑.首先,从图像中分割出叶片,根据叶片距离分离出单个植株;将连续相邻的两幅图像进行拼接获得拼接变换,并将两幅图像中各植株图像与空图像按照获得的拼接变换计算出植株拼接变换图;其次,求出各株拼接变换图的叶片图像质心,采用距离门限的方法判断两植株图像是否为同株作物;在此基础上,采用传递闭包获得各株作物在不同图像中的集合,并用距离中心最小的方法从各株集合中选择出最佳图像,用5种作物对该算法进行验证,结果表明:株数检测准确率达到了100%,提出的方法也成功地将各植株图像精确地提取出来,同时算法显示出高效的计算效率.

A method was constructed to extract individual plant image from the continuously-collected images.The leaf surface region segmentation was obtained,and the individual plant image was separated based on leaf surface image by leaf distance.The two consecutive adjacent images were stitched together,and the newly-formed images were transformed and stitched with empty images to produce individual plant image for determining the centroid.The distance between the centroid of all stitched images in the previous image and the centroid of all the stitched images in the subsequent image was compared.The distance threshold method was used to identify whether the two images were the images of the same crop.Transfer closure was applied for obtaining the plant image set of each crop in multiple continuous images,and the image closest to the center was selected as the best image.Five crops were used to verify the proposed method.The results show that the accuracy of plant number reaches 100%.The proposed method can successfully and precisely extract individual plant image with high computational efficiency.