杂草识别中植被群落特征提取的研究

In the light of fact that the effect of weed recognition in spraying herbicides needs to be improved, it is put forward that the fundamental purpose of image feature extraction in weed recognition is not to distinguish between weeds and crops, but to determine whether to spray herbicides and the spraying dose. According to this hypothesis, it is proposed to use the differences in leaf shape of different plants to represent the feature of the situation of weeds growing with the vegetation community and conduct image feature extraction. As to the extraction of vegetation leaf shape, because there is great difference in the light intensity when there is specular reflection between the leaves in different angles and the soil in the background, it is proposed to use this phenomenon to conduct the study of feature extraction of leaf blade margin of vegetation community. As to variable herbicides spraying in the fields, the law of the change of light intensity and the change of the pixels of the leaf blade margin of vegetation community in the images will be clarified，the law of the change of leaf blade margin of vegetation community of crops with different leaf shape and weeds in different crop-weed mixed growing situation will be revealed, and the representation of the feature of the situation of weeds growing with the vegetation community and related image feature extraction will be explored. This will directly connect the image feature extraction in the filed with herbicide spraying decision and provide new ideas for the study of variable herbicides spraying.

针对变量喷洒除草剂中杂草识别效果有待提高的现状，提出杂草识别中提取图像特征的根本目的不是为了区分杂草和作物，而是为了进行是否喷洒除草剂及确定喷洒剂量的决策。根据这一假设，提出用不同植物叶片形状存在的差异进行表示植被群落杂草伴生情况图像特征的提取研究。而针对植被叶片形状的提取，利用不同生长角度叶片和背景土壤出现镜面反射时的光照强度存在较大差异的现象，进行植被群落叶片边缘特征提取的研究。对于田间变量喷洒除草剂，将阐明光照强度变化与图像中植被群落叶片边缘像素点变化的规律，揭示不同叶形作物与杂草在不同杂生程度下植被群落叶片边缘的变化规律，探索表示植被群落杂草伴生程度图像特征的提取问题。这将把田间图像特征的提取与喷洒除草剂决策直接连接起来，为变量喷洒除草剂的研究提供新的思路。

针对基于田间图像的杂草识别，利用不同形状叶片反光情况存在差异，课题组提出应用植被群落特征描述伴生杂草的生长情况。课题组提出杂草识别的本质是根据伴生杂草情况进行除草剂喷洒的决策，而不以识别作物和杂草为目的。课题的研究内容主要包括4个方面。（1）课题分析了不同光照强度下田间图像的分割问题，研究了光强与田间图像不同分割结果之间的关系，相关成果应用到田间导航线的提取中也取得了较好的效果。（2）基于光强对田间图像分割的影响分析了植被群落叶片边缘的提取，课题组将植被群落叶片边缘特征的提取转换为不同光照强度下田间图像的融合问题。通过基于像素点级的图像融合提出了2种植被群落边缘特征的提取方法，在室内实验和田间实验均取得了一定的效果。相应的边缘提取结果能够明显的区分不同叶片形状的植被群落，这为植被群落特征的提取提供了基础。（3）从植被群落叶片边缘提取方法出发，课题组利用小波变换、形态学操作、感兴趣区域色差、多阈值分割提出了4种植被群落特征提取算法。针对不同数量伴生杂草下植被群落叶片反光情况的变化，相应的植被群落特征描述了不同数量伴生杂草植被群落的叶片边缘在田间图像中像素点数量的差异。通过蔬菜田间实验验证了形态学操作和多阈值分割的方法能够较好的识别是否需要喷洒除草剂。在初步实验中感兴趣区域色差和多阈值分割的方法对于山区茶园的杂草伴生情况能够有较好的识别效果。（4）基于无人机喷洒除草剂进行特征的选择优化。课题组提出了2种无人机喷洒处方图生成算法，并据此进行了不同特征喷洒效果的仿真。总体上，课题组利用光照在不同叶片上的反光情况存在差异，建立了植被群落叶片边缘提取算法，并据此提出了描述伴生杂草数量的植被群落特征，根据这些植被群落特征进行是否需要喷洒除草剂的决策取得了较好的效果。特别是课题研究扩展到无人机平台上，使得相关图像特征更接近生产实际。