基于计算机视觉技术的玉米发育期识别研究

发布时间：2018-05-22 09:40

本文选题：玉米发育期 + 计算机视觉　；参考：《江南大学》2017年硕士论文

【摘要】：科技的进步促使农业的发展迎来了现代农业技术革命,信息、新材料、生物及可持续发展技术综合于一体的农业智能化,为农业生产开辟新的道路。计算机视觉技术作为信息技术的重要分支被广泛应用于农作物精细农业自动化中,特别是生长状态监测及杂草识别等技术,具有非破坏性、检测精度高、效率高、灵活及实时性好的特点,避免了人工观测的主观性强、效率低的弱点。目前计算机视觉技术用于农作物发育期的监测已有一些研究成果,但其检测精度不高且普适性差。因此,为了克服上述缺点,本文利用计算机视觉技术对玉米出苗期、三叶期、七叶期及抽雄期的自动观测算法进行研究,观测记录玉米到达各时期的日期,对指导玉米田间农事活动有重要意义。玉米出苗期自动观测的主要难点是玉米苗很难识别,玉米田间土壤及麦秆等各种干扰较多。因此本文首先利用基于视网膜模型的图像增强算法对玉米出苗期图像进行增强,提高玉米苗的对比度,再利用改进的超绿指数对增强后图像进行灰度化,之后利用迭代式阈值方法分割出玉米苗,最后利用每天平均苗个数的相对变化率来判断玉米是否进入出苗期。玉米三叶期时玉米苗与图像背景对比度低,不易区分。因此本文提出一种玉米苗增强算法,利用玉米三叶期图像的RGB各通道值重新计算G通道值,得到的新的RGB图像中玉米苗与背景差异增大,易于区分。图像增强后再利用Fisher线性判别对图像进行灰度化处理,最后利用二维Otsu算法分割出玉米苗。该方法可以很好地从不同条件下的田间图像中分割出玉米苗。玉米七叶期图像分割算法与玉米三叶期的相同。抽雄期时期本文选择利用颜色特征进行玉米雄穗分割识别,但是不同地区不同品种的玉米雄穗的颜色有些差异,所以本文提出一种玉米抽雄期自动观测算法,该算法首先对玉米雄穗的颜色进行增强,在YCbCr颜色空间中分别对Cb、Cr分量图进行增强,然后再变换到RGB颜色空间中,玉米雄穗被增强了,与背景颜色差异较大,且不同地区不同品种的玉米雄穗颜色相近;增强之后利用改进的Kmeans对玉米雄穗的灰度图像进行聚类分割,利用连通域的面积去除干扰,分割出玉米雄穗,玉米雄穗数量大于设定的阈值即可判断玉米进入抽雄期。河南、河北、山东及内蒙古作为我国玉米生长的主要地区,本文特选取上述四个地区玉米不同发育期的图片进行处理,验证本文所提出的自动观测算法的有效性,并与人工观测结果进行对比,结果表明本文提出的算法可行有效。
[Abstract]:The progress of science and technology promotes the development of agriculture to usher in modern agricultural technological revolution, information, new materials, biology and sustainable development technology integrated into one of the intelligent agriculture, open up a new way for agricultural production. As an important branch of information technology, computer vision technology is widely used in crop fine agriculture automation, especially in the fields of growth state monitoring and weed identification. It has the advantages of non-destructive, high detection accuracy and high efficiency. The features of flexibility and real-time avoid the weakness of subjectivity and low efficiency of artificial observation. At present, computer vision technology has been used in the monitoring of crop development period, but its detection accuracy is not high and its universality is poor. Therefore, in order to overcome the above shortcomings, this paper studies the automatic observation algorithms of maize seedling stage, trefoil stage, heptleaf stage and pumping stage by using computer vision technology, and records the dates of maize arriving at each stage. It is of great significance to guide the field farming activities of maize. The main difficulty of maize seedling automatic observation is that it is difficult to identify maize seedling, and there are many kinds of interference such as maize field soil and wheat stalk. In this paper, firstly, the image enhancement algorithm based on retinal model is used to enhance maize seedling image to improve the contrast of maize seedling, and then the improved super-green index is used to grayscale the enhanced image. At last, the relative change rate of the average number of seedlings per day was used to determine whether the maize entered the emergence stage. The contrast between maize seedling and image background is low and it is difficult to distinguish between maize seedling and image background. Therefore, a corn seedling enhancement algorithm is proposed in this paper. The G-channel value is recalculated by using the RGB channel values of the maize trefoil image. The difference between the maize seedling and the background in the new RGB image is larger, and it is easy to distinguish between the maize seedling and the background. After image enhancement, the grayscale image is processed by Fisher linear discriminant, and finally corn seedling is segmented by two-dimensional Otsu algorithm. This method can segment maize seedlings from field images under different conditions. The image segmentation algorithm is the same as that of maize trefoil stage. In this paper, we choose to use color features to segment and recognize the male ear of maize, but there are some differences in the color of the spikes of different varieties in different regions, so this paper proposes an automatic observation algorithm for the heading period of maize. In this algorithm, the color of maize spikes is enhanced first, and the CBN Cr component map is enhanced in the YCbCr color space, then transformed into the RGB color space, the spikes are enhanced, which is different from the background color. The color of maize spikes of different varieties was similar in different regions, and the improved Kmeans was used to cluster and segment the gray images of maize male ears, and the area of the connected region was removed to segment the male ears of maize. The number of maize spikes was greater than the set threshold to determine the heading stage of maize. Henan, Hebei, Shandong and Inner Mongolia are the main areas of maize growth in China. In this paper, the images of maize at different developmental stages in the four regions mentioned above are selected to verify the effectiveness of the automatic observation algorithm proposed in this paper. The results show that the proposed algorithm is feasible and effective.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S513;TP391.41

【相似文献】