自然光照条件下苹果识别方法对比研究  

Comparison of apple recognition methods under natural light

作　　者：麦春艳 郑立华[1] 肖昌一 李民赞[1] 

机构地区：[1]中国农业大学信息与电气工程学院/现代精细农业系统集成研究教育部重点实验室,北京100083

出　　处：《中国农业大学学报》2016年第11期43-50,共8页Journal of China Agricultural University

基　　金：国家自然科学基金资助项目(31071330)

摘　　要：针对自然光照条件下果园苹果识别效果不佳的问题,从苹果的颜色分割和形状提取2方面进行对比研究,提出一种自然光照条件下的苹果识别方法。利用错检率、漏检率和处理速度3个量化指标综合对比分析颜色阈值、SVM和BPNN 3种苹果颜色分割方法的处理效果。比较6种边缘检测算法对苹果区域图像的边缘检测效果,并使用Hough圆检测算法对苹果形状进行提取,以获得苹果的圆心和半径。试验结果表明:由BPNN的苹果颜色分割方法以及结合Log和Hough的苹果形状提取方法所构建的果实识别算法具有较高的鲁棒性和准确性,能有效克服果实遮挡、重叠和颜色变异等问题,果实平均识别率可达91.6%。To solve the problem of the poor apple recognition performance under the natural light,two approaches of fruit color segmentation and shape extraction were studied to find an effective method for identifying apple fruits under natural light conditions.For the fruit segmentation,color threshold,SVM（Support vector machine）and BPNN（Back propagation neural network）were carried out and the results were compared and analyzed.The results showed that the segmentation result based on neural network was the best because of its high processing speed,low target missing rate and low identifying fallout ratio.For the fruit shape extraction,6edge detection algorithms were applied and compared,the results indicated that Log edge detection algorithm was the best.Hough transform was also used to extract the characteristics of fruit shape from the edge image by circle detection and the center and radius of each fruit.The results showed that the apple identification approach based on BPNN apple color classification method,Log edge detection algorithm and Hough shape extraction algorithm achieved high robustness and accuracy,could effectively overcome the problems such as shading,overlapping and color variation,and its average identification rate reached 91.6%.It illustrated that the recognition algorithm for apple fruits was effective to identify apple fruits from apple tree＇s image taken under natural light conditions.

关 键 词：苹果 阈值分割 BPNN SVM 边缘检测 Hough圆检测 果实识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S216[自动化与计算机技术—计算机科学与技术]