在线式图像的甘蔗长势无损监测研究

发布时间：2018-05-31 01:05

本文选题：甘蔗 + 长势监测　；参考：《浙江大学》2017年硕士论文

【摘要】：农作物的长势监测是农业气象观测的一个重要组成部分,可以为农作物种植田间管理、产量预报提供基础数据。长期以来,作物长势地面监测主要通过人工观测方式采集获取,靠观测人员到田间进行实地观测,这种方法主观性强且费时、费力。随着计算机技术和互联网在农业中的广泛应用,人工观测方法已不能满足现代化农业发展的需求。因此,亟需一种在线、实时、自动的观测方式来实现作物长势的监测。在大田环境下,利用成像设备结合计算机、图像处理、互联网技术对作物进行实时监测是一种可选的方法。本文通过在广西省壮族自治区柳州市柳城县社冲乡的甘蔗田进行实地实验,采用全要素农业自动气象观测系统对获取的甘蔗图像进行提取、分析,获取甘蔗的主要生长特征参数,目的是为甘蔗长势的在线定量化无损监测提供技术基础。通过图像自动采集装置获取2015年和2016年全生育期甘蔗图像,利用图像处理技术对获取的图像进行分析、处理,并结合RGB分量所构建的颜色指数构建了甘蔗叶面积指数估算模型、甘蔗覆盖度估算模型,从而反映甘蔗的长势状况。采用图像分割、图像二值化、连通域标记、连通域特征统计等技术方法实现了对甘蔗出苗期进行自动检测,为甘蔗出苗期的农事管理决策提供了基础数据。通过理论和实验相结合的方法,研究获得了以下主要结果:(1)通过对摄像机正下视图像数据进行预处理,构建了基于颜色指数的甘蔗叶面积指数估算模型,并且通过分析不同颜色指数、时间组合对甘蔗叶面积指数估算模型精度的影响,确定甘蔗叶面积指数最佳估算指数为G-B,且G-B指数采用9点、11点、15点获取的数字照片计算平均值时,估算精度最高,预测R2最大为0.8164,RMSE最小为0.1211。(2)通过对摄像机正下视图像数据进行预处理,采用PFMRF法对甘蔗图像进行分割,由此计算甘蔗的覆盖度作为覆盖度参考值;同时计算图像的九种颜色指数,通过对覆盖度参考值和颜色指数进行相关分析,发现甘蔗的覆盖度和由图像计算的颜色指数之间存在较好的相关性。通过分析、建模、验证,最终得出,以颜色指数ExG-ExR、ExG或CIVE为变量的甘蔗覆盖度估算模型精度均较好。其中,最佳的颜色指数是ExG,利用上午11:00的图像数据所建立的模型估算精度最高,RMSE和MAE分别为0.0484、0.0409。(3)通过对相机图像数据和CCD摄像机图像数据进行分析,以图像上甘蔗幼苗的形状作为特征,将图像中的杂草去除,进而识别出甘蔗幼苗。然后对识别出的甘蔗幼苗的分布特性进行分析,进而判断该图像是到达出苗期。通过与观测人员记录的甘蔗发育期时间比较,发现基于图像的甘蔗出苗期自动识别算法精度较高,误差在±3天以内。
[Abstract]:Crop growth monitoring is an important part of agrometeorological observation, which can provide basic data for crop planting field management and yield forecast. For a long time, crop growth monitoring is mainly acquired by artificial observation, and is observed in the field by observers. This method is subjective, time-consuming and laborious. With the wide application of computer technology and Internet in agriculture, artificial observation method can not meet the needs of modern agricultural development. Therefore, an online, real-time and automatic observation method is needed to monitor crop growth. In the field environment, it is an optional method to use imaging equipment combined with computer, image processing and Internet technology to monitor crops in real time. In this paper, the sugarcane field in Liucheng County, Liuzhou City, Guangxi Province was tested in the field, and the obtained sugarcane image was extracted and analyzed by the automatic meteorological observation system of all-factor agriculture. The main growth characteristic parameters of sugarcane were obtained in order to provide technical basis for on-line quantitative nondestructive monitoring of sugarcane growth. The sugarcane images of the whole growth period of 2015 and 2016 were acquired by the automatic image acquisition device. The obtained images were analyzed and processed by image processing technology, and the estimation model of sugarcane leaf area index was constructed by combining the color index constructed by the RGB component. The estimation model of sugarcane coverage can reflect the growing condition of sugarcane. The techniques of image segmentation, image binarization, connected domain marking and connected domain feature statistics are used to realize automatic detection of sugarcane seedling stage, which provides basic data for sugarcane seedling management decision. By combining theory with experiment, the following main results are obtained: 1) based on color index, a model for estimating sugarcane leaf area index is established by preprocessing camera forward down view image data. The effects of different color indices and time combinations on the precision of the model were analyzed. The best estimation index of sugarcane leaf area index is G-B, and the G-B index is the best when the average value of digital photographs obtained from 9 points and 11 points and 15 points is used to calculate the average value, the estimation accuracy is the highest. The prediction R2 maximum is 0.8164 RMSE minimum 0.1211.02) by preprocessing the camera forward downlooking image data, the sugarcane image is segmented by PFMRF method, the coverage of sugarcane is calculated as the reference value of coverage, and the nine color indices of the image are calculated at the same time. By analyzing the correlation between the reference value of coverage and the color index, it is found that there is a good correlation between the coverage of sugarcane and the color index calculated by image. Through analysis, modeling and verification, it is concluded that the precision of sugarcane mulching estimation model with color index ExG-ExRGExG or CIVE as variable is better. Among them, the best color index is Exg, the model established by using 11:00 image data has the highest estimation accuracy (RMSE and MAE are 0.0484n0.0409.93) by analyzing the camera image data and the CCD camera image data. Taking the shape of sugarcane seedling in the image as the feature, the weed was removed from the image, and then the sugarcane seedling was identified. Then, the distribution characteristics of sugarcane seedlings were analyzed, and the image was determined to reach the emergence stage. By comparing with the time of sugarcane development recorded by the observer, it is found that the automatic recognition algorithm of sugarcane seedling period based on image has a high accuracy and the error is within 卤3 days.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S566.1;S126

【相似文献】