摘要
针对机器人采摘番茄因用途不同而进行选择性收获的问题,对人工选择的番茄图像样本集进行特征分析,从番茄的摄像机透视几何出发,提出将番茄表面红色调所占着色面积比与其他色调所占着色面积比的差值作为描述番茄成熟程度的主要分组特征。利用该特征并结合番茄着色区域整体色调均值和方差,用BP神经网络建立番茄成熟度的判别模型。通过模型测试和噪声水平测试表明,将着色面积比差值和色调均值作为模型的两个输入时,模型的准确判别率和抗干扰能力都是最佳的。模型测试的准确判别率为97.5%,当噪声水平在0.05以下时准确判别率可达到95.26%以上,可以为番茄自动收获作业提供一定的理论参考依据。
Aiming at the problems of selective harvesting-tomato for robot according to various purposes, feature extraction and analysis of tomato image from artificial selection were firstly performed. Proceeding from camera perspective geometry of tomato, difference value between rendering area ratio of red hue to whole tomato region and rendering area ratio of other hue to whole tomato region was presented by way of main grouping feature of judgment model to describe maturity of tomato. An automatic judgment model on maturity of tomato based on BP NN was built up by using above feature combined with hue mean and variance of whole rendering region of tomato. Validation test and noise level test of models show that the model can keep higher accuracy and nicer anti-interference when difference value of area ratio and hue mean are selected as two inputs of BP NN. The accuracy of validation test and noise level test is 97.5%. The accuracy can reach above 95.26% when noise level is under 0.05. The model can provide theoretical reference of automatic harvesting-tomato for robot.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2009年第10期146-150,168,共6页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家自然科学基金资助项目(60575020
50805067)
中国博士后科学基金资助项目(20080441073)
关键词
农业机器人
番茄
计算机视觉
神经网络
成熟度
Agricultural robot, Tomato, Computer vision, Neural network, Maturity
