基于剪切波变换和无人机麦田图像的区域杂草识别方法

发布时间：2018-04-05 15:11

本文选题：无人机　切入点：图像识别　出处：《农业工程学报》2017年S1期

【摘要】：区域杂草的识别有利于植保作业中的除草剂精准喷施。现有图像处理技术主要针对行间和株间杂草,而传统的图像采集与分析设备对苗期麦田杂草的识别存在一定局限性,难以满足非人工的区域性喷洒农药等作业需求。由于麦田区域中的麦苗和杂草具有形态和颜色区分度差的特点,传统的图像识别方法难以有效识别。针对此问题,该文提出利用剪切波变换对无人机麦田区域图像中杂草进行识别。该方法利用其自身的方向敏感性以及在纹理识别中的方向无关性,根据麦田区域图像在杂草较多的部分叶片纹理杂乱,反之则纹理相对规则的特点,处理得到不同尺度和不同方向下小麦与杂草的剪切波系数。然后针对小麦和杂草剪切波系数的不同特征,对剪切波系数矩阵进行归一化处理,同时对其均值和方差进行了统计分析,得到麦苗和杂草剪切波系数图中竖直锥第二尺度所有系数均值的区分值约为0.07,第二尺度各个方向的均方差均值的区分值约为0.08。通过对含杂草麦苗区域图像以及麦苗区域图像的验证,准确率为69.2%,效果优于传统的灰度共生矩阵方法。此外,该文对无人机拍摄区域图采用分块的方法,实现了对非麦苗区域的有效标识。由此可见,剪切波变换方法能够为基于低空植保无人机喷洒农药中的区域杂草识别提供参考。
[Abstract]:The identification of regional weeds is beneficial to the precise spraying of herbicides in plant protection operations.The existing image processing technology is mainly aimed at interrow and interplant weeds, but the traditional image acquisition and analysis equipment has some limitations on weed identification in seedling wheat field, and it is difficult to meet the needs of non-artificial regional spraying pesticides.Because the wheat seedlings and weeds in wheat field have the characteristics of poor morphological and color differentiation, the traditional image recognition method is difficult to identify effectively.In order to solve this problem, a shear wave transform is proposed to identify weeds in wheat field images of unmanned aerial vehicles (UAVs).Based on the orientation sensitivity and direction-independence in texture recognition, the method is based on the texture clutter of some leaves with more weeds in the wheat field image, whereas the texture is relatively regular.The shear wave coefficients of wheat and weeds were obtained under different scales and directions.Then, according to the different characteristics of shear wave coefficients of wheat and weeds, the shear wave coefficient matrix is normalized, and its mean and variance are statistically analyzed.The difference value of all mean values of all coefficients in the second scale of vertical cone is about 0.07, and that of the mean mean of mean mean of mean square deviation of every direction of the second scale is about 0.08 in the shear wave coefficient map of wheat seedlings and weeds.The veracity of the image of wheat seedling region containing weeds and that of wheat seedling area is 69.2, which is better than that of the traditional gray level co-occurrence matrix method.In addition, the region map of UAV is divided into blocks and the effective identification of non-wheat seedling area is realized.Therefore, the shear wave transform method can provide a reference for the identification of regional weeds in sprayed pesticides based on low altitude plant protection UAV.
【作者单位】：中国农业大学现代精细农业系统集成研究教育部重点实验室;中国农业大学信息与电气工程学院;北京邮电大学世纪学院移动媒体与文化计算北京市重点实验室;
【基金】：国家自然科学基金资助项目(31301240) 北京市自然科学基金资助项目(4172034)
【分类号】：S451;TP391.41

【相似文献】