BP神经网络在奶牛体强判别中的应用研究

根据专家知识从奶牛图象中提取了描述奶牛体强的１２个特征参数；设计了一个三层前向神经网络，选择了一组奶牛图象训练样本。网络训练试验表明，本网络收敛性好且学习次数少；对现场采集的１０头奶牛图象进行奶牛体强判别，判别结果和人工评定的结果基本一致；对全部输人特征参数进行了主变量分析，给出了采用不同数量特征参数的优化组合方案。作者认为神经网络可以取代人工评定奶牛体强。

Study on Appraisal of the Strength of Dairy Cow's Body Using BP Neural Networks

ccording to the twelve characteristic parameters describing the strength of Dairy cow's body have been extracted from cow's image based on experts'knowledge a three-layer BP neural network has been designed and a set of training samples have been selected. The training tests of the network showed that the network has a good convergence and requires less learning tines. The appraising results for 10 cow's images by the neural network are comparable with that of by hand. The optimum combinations adopting different numbers of characteristic parameters have been obtained by analyzing the main variables of all the input parameters. Author believes that neural networks can replace manual appraisal of the strength of cow's body.