寒地大豆病虫害诊断方法研究

发布时间：2018-06-25 14:05

本文选题：大豆病虫害诊断 + 层次分析法　；参考：《东北农业大学》2017年硕士论文

【摘要】：在我国专业农业诊断知识的匮乏和专业农业专家的稀缺之间的矛盾已经日益严重,也严重阻碍了我国农业精准化、现代化的发展,而解决这一矛盾的重要渠道,就是实现农业的智能化。在我国众多大豆种植区域中病虫害的存在往往会造成10%以上的直接经济损失,个别地区会达到30%以上,多种形式的病虫害已经极大的制约我国出产的大豆的产量及品质。目前人工智能技术已广泛应用于疾病诊断领域,人工神经网络在作物病虫害诊断中的应用已成为一种流行趋势。因此本文拟针对大豆病虫害进行精准判定,选取模糊神经网络进行模型建立,并引入AHP层次分析法自动生成和调整隶属度函数,探讨结合模糊神经网络与AHP层次分析法进行病虫害诊断的可行途径。通过仿真实验显示,利用模糊神经网络与层次分析法相结合的模型用于大豆病虫害诊断具有泛化能力强、诊断速度快、正确率高等优点,不失为一个好的选择。具体内容如下:首先,输出采用7种我国具有代表性的食心虫等虫害。对182个大豆虫害样品,依据危害方式、危害症状等8种性状进行诊断,选择136个大豆虫害样本作为训练集,并用46个样本作为测试集。通过对大豆病虫害症状的收集整理和分析,分别使用对输入/输出向量进行数字化编码和对输入使用AHP层次分析法,将用两种方法处理后的数据用作神经网络的输入向量。其次,分别建立3种用于训练和仿真的神经网络模型。分析BP神经网络中的最佳隐含层节点数、训练目标、学习速率和训练次数等参数对网络性能的影响;论证RBF径向基神经网络中径向基密度参数对训练结果的影响;同时论证模糊神经网络中隐层节点数和训练次数等参数对模型的响应结果。最后,对比三种类型神经网络对不同大豆病虫害进行诊断后的准确率,从实验结果论证了模糊神经网络应用于大豆病虫害具有最佳诊断效果。实验结果表明,选择层次分析法对输入进行处理以及模糊神经网络进行建模,在46个测试样本中,共有44个样本进行了预测,识别率高达95%,证明了该方法对大豆害虫的判别是可行的。
[Abstract]:The contradiction between the lack of specialized agricultural diagnostic knowledge and the scarcity of specialized agricultural experts in China has become increasingly serious, which has also seriously hindered the development of agricultural precision and modernization in our country. It is to realize the intelligence of agriculture. In many soybean planting areas in China, the existence of pests and diseases will often cause more than 10% of direct economic losses, and a few areas will reach more than 30%. Many forms of diseases and pests have greatly restricted the yield and quality of soybean produced in China. At present, artificial intelligence technology has been widely used in the field of disease diagnosis, and the application of artificial neural network in the diagnosis of crop diseases and insect pests has become a popular trend. Therefore, this paper intends to accurately judge soybean pests and diseases, select fuzzy neural network to establish the model, and introduce AHP to automatically generate and adjust membership function. This paper discusses the feasible ways to diagnose diseases and insect pests by combining fuzzy neural network with AHP. The simulation results show that the model combined with fuzzy neural network and analytic hierarchy process has the advantages of strong generalization ability, fast diagnosis speed and high accuracy, and it is a good choice. The specific contents are as follows: first, the output uses 7 representative insect pests and other pests in China. Based on the diagnosis of 182 soybean pest samples, 136 soybean pest samples were selected as training set and 46 samples were used as test sets. By collecting and analyzing the symptoms of soybean diseases and insect pests, the input / output vector was digitalized and the input was analyzed by AHP. The data processed by two methods was used as the input vector of neural network. Secondly, three neural network models are established for training and simulation. The effects of the optimal number of hidden layer nodes, training target, learning rate and training times on the performance of the neural network are analyzed, and the effects of radial basis function density parameters on the training results in RBF radial basis function neural network are demonstrated. At the same time, the response results of the parameters such as the number of hidden layer nodes and the number of training times to the model in the fuzzy neural network are discussed. Finally, the accuracy of three types of neural networks for the diagnosis of different soybean pests and diseases was compared, and the best diagnostic effect of the fuzzy neural network applied to soybean diseases and insect pests was demonstrated from the experimental results. The experimental results show that the analytic hierarchy process (AHP) is selected to deal with the input and the fuzzy neural network is used to model the model. Out of 46 test samples, 44 samples are predicted. The recognition rate is as high as 95%, which proves that this method is feasible for the identification of soybean pests.
【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S435.651;TP183

【参考文献】