水田田埂边界支持向量机检测方法  

Detection Method of Boundary of Paddy Fields Using Support Vector Machine

作　　者：蔡道清 李彦明[1] 覃程锦[1] 刘成良[1] CAI Daoqing;LI Yanming;QIN Chengjin;LIU Chengliang(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院

出　　处：《农业机械学报》2019年第6期22-27,109,共7页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家重点研发计划项目(2016YFD0700505)

摘　　要：提出了基于支持向量机的水田田埂边界线的检测算法。采用支持向量机分类算法代替传统的图像分割算法,分割水田图像,提高了在不同光照条件下田埂边界检测的鲁棒性。图像预处理阶段引入超像素分割算法,大大减少了后续图像处理的计算量,并为支持向量机的模型训练提供大量的样本。选取足够数量的超像素样本,提取其颜色特征和纹理特征,构成19维的特征向量,并作为训练支持向量机模型的输入。使用训练好的支持向量机模型识别新图像中的水田田埂区域,模型评价指标F1分数达到90.7%。采用霍夫变换提取田埂边界,在NVIDIA的Jetson TX2硬件平台上,算法总运行时间在0.8s以内,有效满足了水田直播机的实时性要求。Automatic navigation is the core elements of agricultural intelligence,and the machine vision based navigation route detection is the core content of automatic navigation system. An algorithm based on support vector machine was proposed to detect paddy field boundary. Support vector machine,instead of traditional image segmentation algorithms,was used to segment the paddy field image,and the robustness of boundary detection under different illumination conditions was improved. Superpixel segmentation algorithm was used to obtain superpixels instead of pixels for subsequent image processing. Superpixels reduced the computational complexity and provided a large number of samples for model training of support vector machine. A sufficient number of superpixel samples were selected for extracting color features and texture features to form a 19-dimensional feature vector. Color features were statistical properties in RGB and HSV color spaces,including R average,G average,B average,H average,S average,V average,H variance,S variance and V variance. Texture features included gradient amplitude mean and weighted gradient direction histogram. Then support vector machine model was trained and used to identify the paddy ridge field in the new picture. In order to judge the performance of the algorithm,the superpixel classification results and the actual manual labeling results were compared based on the 50 images containing paddy ridge field. The recognition F1-score can reach 90. 7%.Finally,Hough detection was used to extract the boundary of the paddy ridge field. It took less than 0. 8 s on NVIDIA’s Jetson TX2 hardware platform by the algorithm and can meet the real-time requirement of agricultural machinery.

关 键 词：田埂边界 机器视觉 支持向量机 霍夫检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]