葡萄病害诊断BP网络模型关键问题研究

BP Network Model for Grape Disease Diagnosis Research on Key Issues

针对葡萄病害诊断BP网络模型建立和应用中的领域知识编码方法、BP网络基本要素和网络应用等关键问题提出了解决方案。研究了以二进制编码为理论基础上的动态编码方法,对知识库实际存储的诊断参数进行实时编码,累计诊断参数编码位数作为输入层神经元个数,采取"n中取1"的策略确定网络输出。通过定义症状相似系数,反映实际诊断症状与训练样本症状的相似程度。上述方法应用在葡萄常见的16种病害诊断中,取得了满意效果。

We proposed solution strategy for some key issues in diagnosis of grape diseases including the domain knowledge encoding method in modeling and applying BP network to the diagnosis of grape diseases, the basic factor of BP network and web applications. We discussed on dynamic encoding method based on binary encoding theory as well as real - time coding for diagnosis parameters exist in knowledge base. We set the number of the neurons in input layer by accumulating the bit number of code of the diagnosis parameters while the “1 from N” is considered as the strategy for the network output. We use the definition of the similarity coefficient to reflect the similarity between the symptom of diagnosis and training samples. By applying our method to the diagnosis of familiar 16 diseases of grasp, we have yielded satisfactory results.