基于冠层颜色特征的大豆缺素症状识别研究

发布时间：2018-07-09 19:06

本文选题：冠层图像 + 颜色特征　；参考：《西北农林科技大学学报(自然科学版)》2016年12期

【摘要】：【目的】针对寒地大豆发生缺素症状时冠层颜色变化复杂性,建立基于冠层图像颜色特征的大豆缺素症状识别新方法。【方法】采用无土盆栽试验,以垦农18为供试大豆品种,设计缺氮、缺磷、缺钾3种营养状况,采集大豆缺素症状的冠层图像样本,利用图像灰度直方图结合主成分分析方法,提取大豆冠层图像的红光值R、绿光值G、蓝光值B,计算最佳颜色特征蓝光标准化值B/(R+G+B)和绿光标准化值G/(R+G+B),将其作为正则化模糊神经网络输入向量,并利用实数编码的遗传算法改进传统梯度下降学习算法,将其作为模糊神经网络的学习方法,同时应用传统梯度下降算法和改进梯度下降算法训练神经网络参数并比较。【结果】应用遗传计算改进的梯度下降学习算法计算时,迭代次数为277次,其各项计算指标均明显优于传统梯度下降算法,大豆缺素症状识别准确率达100%;而采用传统的多元线性回归方程和BP神经网络算法计算时,识别准确率分别为52.50%,68.33%。【结论】以大豆冠层图像颜色特征为基础,利用改进学习算法的神经网络模型,能够快速有效地挖掘出大豆缺素症状与颜色特征向量之间的模糊逻辑映射关系,为大豆缺素症状识别提供了一种快速且准确的方法。
[Abstract]:[objective] in view of the complexity of color change of canopy in cold region soybean, a new method based on color characteristics of canopy image was established to identify soybean deficiency symptoms. [methods] Soil-free pot experiment was used to test soybean cultivar Kannong 18. Three nutrition conditions were designed: nitrogen deficiency, phosphorus deficiency and potassium deficiency. The canopy image samples of soybean deficiency symptoms were collected, and the image gray histogram combined with principal component analysis (PCA) was used. Red light value R, green light value G, blue light value B of soybean canopy image were extracted, and the best color characteristics, blue light standardization value B / (R G B) and green light standard value G / (R G B) were calculated as input vectors of regularized fuzzy neural network. The traditional gradient descent learning algorithm is improved by real-coded genetic algorithm, which is regarded as the learning method of fuzzy neural network. At the same time, the traditional gradient descent algorithm and the improved gradient descent algorithm are used to train and compare the parameters of the neural network. [results] in the computation of the improved gradient descent learning algorithm based on genetic computation, the number of iterations is 277 times. All the calculation indexes are obviously superior to the traditional gradient descent algorithm, and the accuracy of soybean deficiency symptom recognition is 100, while the traditional multivariate linear regression equation and BP neural network algorithm are used to calculate, [conclusion] based on the color characteristics of soybean canopy image, the neural network model of improved learning algorithm is used. The fuzzy logic mapping relationship between soybean deficiency symptoms and color feature vectors can be quickly and effectively mined, which provides a fast and accurate method for soybean deficiency symptom recognition.
【作者单位】：黑龙江八一农垦大学信息技术学院;中国农业大学信息与电气工程学院;黑龙江八一农垦大学农学院;
【基金】：黑龙江省自然科学基金项目(QC2016031) 黑龙江省大学生创新创业训练计划项目(1022320169433) 黑龙江省教育厅科学技术研究项目(12521375)
【分类号】：S435.651;TP391.41

【相似文献】