基于BP模型的黄瓜霜霉病图像特征识别系统的设计  

Design on Image Features Recognition System of Cucumber Downy Mildew Based on BP Algorithm

作　　者：张大伟[1,2] 王军[2] 

机构地区：[1]东北大学机械工程与自动化学院,辽宁沈阳110004 [2]沈阳理工大学信息科学与工程学院,辽宁沈阳110168

出　　处：《沈阳建筑大学学报（自然科学版）》2009年第3期574-578,共5页Journal of Shenyang Jianzhu University：Natural Science

基　　金：863计划项目(2003AA41250);辽宁省科技厅计划项目(2007219010);辽宁省教委A类基金项目(20243303)

摘　　要：目的设计农作物黄瓜霜霉病叶部图像特征识别系统.方法应用神经网络技术的反向传播算法,利用基于BP模型的神经网络进行了黄瓜霜霉病叶部的图像特征提取与模式识别生成人工神经网络的结构图.通过黄瓜霜霉病不同阶段110个样本进行训练,并使用另外100个样本进行测试.结果得到最优的隐含节点数,确定该算法参数的最优解,其中包括惯性项系数α、权值修正系数η与各层网络的权系数.样本训练在有限的允许次数内收敛至误差小于规定的数值并得到次数上最优的隐含节点数目.BP神经网络系数收敛并可达到85%以上的正确识别率.结论提取了黄瓜霜霉病叶部图像的特征,获得正确的分类识别规则,从而进行图像识别工作.BP神经网络用于作物病害诊断的专家识别系统有借鉴意义.The image features recognition system of the leaf segment on cucumber with downy mildew is designed which is used to diagnosis the different course of disease. Using BP algorithm based on neural network technology, the structural diagram of the neural network is generated, the image features of cucumber with downy mildew is extracted and the effectual classification rules are designed. Thus the recognition of downy mildew on cucumber is done. The 110 samples of Cucumber Downy Mildew in different courses are trained and the other 100 samples are tested. The optimum relation of the algorithm＇ s parameters are obtained, including inertial term coefficient a and weights-revising coefficient η and the weights values of each layers. The coefficients of BP algorithm converges can be done and the correct recognition rata is above 85 %. Within the permitted limitation and finite times, the error converges are less than specified numerical and the optimalizing numbers of hidden nodes are got. By extracting image features of cucumber with downy mildew and getting the correct rules of classification, the image recognition can be carried out. The BP neural networks have significance on recognition systems of crops＇ disease diagnosis.

关 键 词：反向传播算法 图像特征 神经网络 黄瓜霜霉病 BP模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]