基于深度信念网络的苹果霉心病病害程度无损检测

发布时间：2018-09-02 06:31

【摘要】：针对现有霉心病无损检测只能检测出有无病害,无法对病害程度进行判断的问题,研究并提出一种基于深度信念网络(deep belief net,DBN)的无监督检测模型。该模型由多层限制玻尔兹曼机(restricted Boltzmann machine,RBM)网络和1层反向传播(back propagation,BP)神经网络组成,RBM网络实现最优特征向量映射,输出的特征向量由BP神经网络对霉心病病害程度分类。对225个苹果样本在波长200~1 025 nm获取其透射光谱后,根据腐烂面积占横截面比例将霉心病害程度分为健康、轻度、中度和重度4种,分别用150个和75个样本作为训练集和测试集,以全光谱数据和基于连续投影算法提取的特征波长数据为输入构建病害程度判别模型,并比较DBN模型与偏最小二乘判别分析、BP神经网络和支持向量机模型的识别效果,实验结果表明,DBN模型病害判别准确率达到88.00%,具有较好的识别效果。
[Abstract]:In order to solve the problem that nondestructive detection of mildew heart disease can only detect disease and can not judge the degree of disease, a kind of unsupervised detection model based on deep belief network (deep belief net,DBN) is studied and proposed. The model is composed of multi-layer restricted Boltzmann machine (restricted Boltzmann machine,RBM) network and layer 1 back propagation (back propagation,BP) neural network to realize the optimal feature vector mapping. The output eigenvector is classified by BP neural network to the degree of mildew heart disease. The transmission spectra of 225 apple samples were obtained at wavelength 2001025 nm. According to the proportion of rotting area to cross section, four kinds of heart disease were classified as healthy, mild, moderate and severe. 150 and 75 samples were used as training set and test set, respectively. Based on the full-spectrum data and the characteristic wavelength data extracted from continuous projection algorithm, the disease degree discriminant model is constructed, and the recognition effect of DBN model and partial least squares discriminant analysis (PLS) is compared with that of BP neural network and support vector machine model. The experimental results show that the accuracy rate of disease discrimination of DBN model is 88.00, and it has better recognition effect.
【作者单位】：西北农林科技大学机械与电子工程学院;西北农林科技大学网络与教育技术中心;
【基金】：国家高技术研究发展计划(863计划)项目(2013AA10230402) 国家自然科学基金面上项目(61473235) 陕西省重大农技推广服务试点项目(2016XXPT-05)
【分类号】：S436.611;TP18

【相似文献】