基于视觉图像的田间甘蓝计数  

Field Cabbage Counting Based on Visual Image

作　　者：宋雨轩 江田 

机构地区：[1]重庆师范大学计算机与信息科学学院,重庆 [2]重庆市数字农业服务工程技术研究中心,重庆

出　　处：《计算机科学与应用》2022年第6期1610-1622,共13页Computer Science and Application

摘　　要：田间蔬菜计数是预估产量的重要技术手段,可以帮助农民提前规划销售、仓储和运输,提高收益。本文以甘蓝为例,实现基于视觉图像的田间蔬菜识别计数。算法是:首先对图像b通道高斯平滑滤波,然后应用b通道直方图实现蔬菜与非蔬菜的自适应阈值分割,再利用改进极限腐蚀算法对分割出的蔬菜二值图像腐蚀,最后用动态生成腐蚀核划分连通域实现蔬菜计数。航拍甘蓝图像的分割、计数实验结果显示:本文算法分割蔬菜与非蔬菜的精度为82.41%,高于OTSU对比算法;本文算法计数准确率达100.00%,召回率为96.08%,F1-score为0.98。实验结果表明,算法是有效的。Field vegetable counting is an important technical means to estimate yield, which can help farmers’ plan sales, storage and transportation in advance and improve income. Taking cabbage as an example, this paper realized the recognition and counting of field vegetables based on visual image. The algorithm is: firstly, the image b-channel Gaussian smoothing filter was used, then the b-channel histogram was used to realize the adaptive threshold segmentation of vegetables and non-vegetables, then the segmented binary image of vegetables was corroded by the improved limit corrosion algorithm, and finally, the connected domain was divided by dynamically generated cor-rosion core to realize the counting of vegetables. The experimental results of aerial cabbage image segmentation and counting showed that the accuracy of this algorithm is 82.41%, which is higher than that of OTSU comparison algorithm;the counting accuracy of this algorithm is 100.00%, the recall rate is 96.08%, and the F1-score is 0.98. Experimental results show that the algorithm is effective.

关 键 词：蔬菜识别 自适应阈值分割 视觉图像 形态学操作 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]