基于CNN的农作物机载高光谱影像质量评价——以棉花为例  

Airborne hyperspectral image quality evaluation of crops based on CNN——in the case of cotton

作　　者：刘汉青 赵庆展[2,3,4] 田文忠 王学文 LIU Hanqing;ZHAO Qingzhan;TIAN Wenzhong;WANG Xuewen(College of Mechanical and Electrical Engineering,Shihezi University,Shihezi,Xinjiang 832002,China;College of Information Science and Technology,Shihezi University,Shihezi,Xinjiang 832002,China;Geospatial Information Engineering Research Center,Xinjiang Production and Construction Crops,Shihezi,Xinjiang 832002,China;Xinjiang Production and Construction Corps Industrial Technology Research Institute,Shihezi,Xinjiang 832002,China)

机构地区：[1]石河子大学机械电气工程学院,新疆石河子832002 [2]石河子大学信息科学与技术学院,新疆石河子832002 [3]兵团空间信息工程技术研究中心,新疆石河子832002 [4]兵团工业技术研究院,新疆石河子832002

出　　处：《石河子大学学报（自然科学版）》2022年第4期510-519,共10页Journal of Shihezi University(Natural Science)

基　　金：中央引导地方科技发展专项资金项目(201610011);新疆生产建设兵团科技计划项目(2017DB005)。

摘　　要：机载高光谱遥感作为近地遥感观测方法中的新方式,对判断作物长势、监测环境状态等农业应用提供了新的技术手段。机载影像数据获取受光照条件、飞行状态等因素干扰,其数据质量对科学研究数据分析结果的可靠性具有重要影响,机载高光谱影像数据的快速质量评价问题亟待解决。本研究提出一种适用于机载棉花高光谱影像的卷积神经网络影像质量分级方法。先获取试验区域内的棉花高光谱影像,通过模拟采集过程中产生的退化进行数据增强;经专家目视评估进行影像判读分级定标,将高光谱数据按比例分为训练集、验证集和测试集;利用TensorFlow框架建立卷积网络模型并进行模型微调与精度验证;最后结合模型评估指标与实验分类结果给出评价。实验结果表明:卷积神经网络影像质量分类评价取得了较好的效果,样本分类准确率达到99.06%,多类别平均分类精确度达到99.07%,Kappa系数为0.9887。研究结果表明采用卷积神经网络对机载棉花高光谱影像质量评价有效可靠,可为机载高光谱农作物影像质量评价提供参考。As a new method of ground-based remote sensing observation methods,airborne hyperspectral remote sensing provides new technical means for agricultural applications such as judging crop growth and monitoring environmental conditions.The acquisition of airborne image data is disturbed by factors such as illumination conditions and flight status.The data quality has an important impact on the reliability of scientific research data analysis results.The problem of rapid quality evaluation of airborne hyperspectral image data needs to be solved urgently.This research proposes a convolutional neural network image quality classification method suitable for airborne cotton hyperspectral images.Firstly,the hyperspectral images of cotton in the research area were obtained,and the data is enhanced by simulating the degradation in the process of image acquisition to avoid the difficulty of convergence due to insufficient samples.Secondly,the hyperspectral data are divided into training set,verification set and test set.In addition to using the TensorFlow framework to build a convolutional network model,model adjustments and accuracy verification were performed.Finally,the evaluation is given by combining model evaluation indicators and experimental classification results.The experimental results show that the image quality classification of convolutional neural network has achieved good results,the sample classification accuracy rate reaches 99.06%,the multi-category average precision reaches 99.07%,and the Kappa coefficient is 0.9887.The results show that the convolution neural network is effective and reliable in evaluating the quality of airborne cotton hyperspectral images,which can provide a reference for the evaluation of airborne hyperspectral crop image quality.

关 键 词：高光谱图像 图像质量评价 卷积神经网络 深度学习 无人机遥感 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP751[自动化与计算机技术—控制科学与工程]