摘要
为解决人工识别种蛋受精误差和效率低等问题,本文提出一种基于DSP硬件平台和遗传神经网络的识别系统。系统以DM6437处理器搭建硬件处理平台,系统识别软件是遗传神经网络算法。它通过拍摄装置获取200枚种蛋图像,图像信息输入DSP系统的处理算法,提取图像的色调(H)分量颜色特征,使用主成分分析法找出色度分量中4个主成分特征,他们的总贡献率超过90%。最后利用遗传神经网络算法输入4个色度主成分特征,预测输出是孵化种蛋的成活性。训练神经网络并用测试集样本验证神经网络。实验结果是遗传神经网络和BP神经网络检测正确识别率分别是93%和86%,表明遗传神经网络的准确率较高,可以实现自动检测种蛋受精和成活。
In order to solve the problems of fertilization error and low efficiency in the artificial identification for breeder eggs,the identification system based on DSP and genetic neural network was proposed.The DM6437 processor was used to set up the processing hardware platform,and the system software was a genetic algorithm neural network.First,200 images were got using machine vision system,and then DSP system was input to extract the hue component color features.Four main component characteristics of the hue component were found out by using principal component analysis(PCA),which total contribution rate was more than 90%.Neural network was used to input the four main component characteristics,and the neural network output is the hatching eggs survival condition.Finally,the neural network was trained and verified,and the correct recognition rate of the genetic neural network and BP neural network is 93% and 86% respectively.The improved neural network can realize automatic detection accurately.
出处
《中国农机化学报》
2015年第3期191-193,209,共4页
Journal of Chinese Agricultural Mechanization
基金
吉林农大青年启动基金(201227)--基于DSP图像处理的鸡蛋品质无损分级研究
吉林农大青年启动基金(201327)--吉林省农产品加工业专利竞争态势分析及技术创新能力研究
吉林省教育厅"十二五"规划课题(201356)--农田大规模传感器网络数据管理技术研究
