基于ResNeXt单目深度估计的幼苗植株高度测量方法  

Method for measuring seedling height based on ResNeXt monocular depth estimation

作　　者：宋磊 李嵘 焦义涛 宋怀波[1,2,3] Song Lei;Li Rong;Jiao Yitao;Song Huaibo(College of Mechanical and Electronic Engineering,Northwest A&F University,Yangling,712100,China;Key Laboratory of Agricultural Internet of Things,Ministry of Agriculture and Rural Affairs,Yangling,712100,China;Shaanxi Key Laboratory of Agricultural Information Perception and Intelligent Services,712100,China)

机构地区：[1]西北农林科技大学机械与电子工程学院,杨凌712100 [2]农业农村部农业物联网重点实验室,杨凌712100 [3]陕西省农业信息感知与智能服务重点实验室,杨凌712100

出　　处：《农业工程学报》2022年第3期155-163,共9页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家重点研发计划(2019YFD1002401);国家自然科学基金项目(31701326);国家高技术研究发展计划(863计划)项目(2013AA10230402)。

摘　　要：幼苗高度是幼苗培育过程中的重要性状,是幼苗生长状况和优良性状筛选的重要参考指标。针对目前研究多选用专业测量工具、使用带有标记的测量手段这一现状,该研究提出了一种基于单目图像深度估计技术的幼苗高度无参测量方法。首先以NYU Depth Dataset V2深度数据集为基础,以ResNeXt 101网络为深度估计网络主体实现植株图像深度估计。通过深度信息计算出拍摄点到植株的真实距离,结合图像中幼苗植株的像素高度和标定好的视场角实现幼苗高度的测量。为验证该方法的有效性,通过采集不同距离下的番茄幼苗图像1728幅,辣椒幼苗图像160幅,甘蓝幼苗图像160幅进行植株高度测量试验。试验结果表明,在拍摄距离为105cm内番茄幼苗平均绝对误差(Mean Absolute Error,MAE)为0.569 cm,均方根误差(Root Mean Square Error,RMSE)为0.829 cm,平均植株高度比例为1.005。辣椒,甘蓝幼苗的MAE为0.616和0.326cm,RMSE为0.672和0.389cm。每株幼苗高度的平均计算时间为2.01s。试验结果表明该方法具有较好的可行性和普适性。不同光照强度下植株高度测量结果表明,在感光度小于160时,植株高度测试结果的MAE为0.81 cm,仍具有较好的测量准确度。当单幅图像中植株个数处于5以内时,MAE和RMSE的平均值分别为0.652和0.829cm。研究结果表明,该模型可以较准确地从单幅图像中检测出多株植株高度,且在不同距离和一定光照强度变化内均可完成多种幼苗植株高度的精确测量。可为幼苗培育和成长时期判断等研究提供一种无损的植株高度测量方法。Seedling height is an important feature in the process of seedling cultivation,and it is also an important reference index for seedling growth and screening of excellent features.In view of the problem that professional measurement tools and marked measurement methods are mostly used in the current research,a measurement method of the seedling height based on monocular image depth estimation technology was proposed in this study.Firstly,the NYU Depth Dataset V2 depth dataset was enhanced to make the model have better expression ability.The depth estimation network structure is a U-shaped network structure,which is divided into encoder and decoder.The encoder part took ResNeXt 101 network as the main body to extract the depth feature information of plant image.The decoder was mainly based on the up sampling,and a jump connection module was added between the encoder and the decoder to increase the detail information of the depth image.Compared with different depth estimation models,the depth estimation model achieved the best Root Mean Square Error(RMSE),which was 0.165.It showed that the depth estimation model can better complete the estimation task of depth information.Through the calibration of the maximum depth value,the real distance from the shooting point to the plant can be calculated according to the depth information,and the seedling height can be measured in combination with the pixel height of the seedling plant in the image and the calibrated field angle.In order to verify the effectiveness of this method,we collected 1728 images of tomato seedlings,160 images of pepper seedlings and 160 images of cabbage seedlings at different distances for plant height measurement.The results showed that within the shooting distance of 105 cm,the Mean Absolute Error(MAE)of tomato seedlings was 0.569 cm,the RMSE was 0.829 cm,and the average plant height ratio was 1.005.The MAE of pepper and cabbage seedlings were 0.616 and 0.326 cm,and the RMSE were 0.672 and 0.389 cm.The average calculation time of the height of each

关 键 词：幼苗植株 高度测量 单目图像 深度估计 ResNeXt 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]