摘要
为使机械手更准确地抓取桃子,提出一种在自然光照条件下识别树上桃子生长形态的方法。首先在5种颜色空间中利用BP神经网络找出识别率最高、误分率最低的颜色特征组合(H,Cr(YCg Cr),Cr(YCb Cr),R-G,2R-G,Cb-Cr),并使用改进的K-means聚类算法实现图像分割;然后利用桃子生长的形态参数(复杂度、延伸率、紧密度等)使用支持向量机分类器进行分类。实验结果表明:对于晴天拍摄的图片,其识别率可达到87.5%;对于阴天拍摄的图片,其识别率可达80.5%。该方法具有一定的实用价值。
In order to pick peaches accurately during the tedious process of harvesting, a recognition method of peaches growth morphology in natural scene method is put forward for robot. In five color spaces, such as H, Cr（ YCgCr）, Cr（ YCbCr）, R - G ,2R - G and Cb - Cr, a color combination that has the lowest recognition error rate is found out based on BP neural network and the improved K-means clustering algorithm is used to segment image. According to peach morphology features, such as complexity, elongation, ec- centricity, etc. , the peach growth morphologies are classfied with support vector machine. Experiment results show that the recognition rate of pictures taken in fine day arrives at 87.5% , and the recognition rate of pictures taken in cloudy day reaches to 80.5%. The re- sults show that the proposed method is practical .
出处
《西华大学学报（自然科学版）》
CAS
2015年第2期6-9,15,共5页
Journal of Xihua University:Natural Science Edition
基金
国家"863"计划项目(2006AA10Z259)
