基于近红外图像的杂草识别研究

Research on Weed Recognition Based on Near-infrared Images

利用近红外图像识别杂草,使用均值法和最大方差自动取阈值法去除土壤背景,利用作物和杂草的形态差异识别杂草。结果表明,利用近红外图像进行土壤背景分割,相对误差小于0.049 0;应用形态学方法可有效识别作物和杂草,识别精度为85.0%-98.8%。

Using near infrared images to recognize weed, the soil background was removed by mean method and maximum variance auto-select thresholds method, and the weeds were recognized according to the morphological difference of crop and weed. The results showed that the relative error of segmentation on soil background by near-infrared images was less than 0. 049 0. And the weed could be effectively recognized with morphological method, and the recognition accuracy was 85.0% -98.8%.