基于多特征融合的玉米前期图像的旱情识别

发布时间：2018-01-19 11:41

本文关键词： 图像处理旱情识别多特征融合遗传算法　出处：《激光与光电子学进展》2017年08期 　论文类型：期刊论文

【摘要】：为实现对玉米植株旱情的分析,针对目前农业干旱指标涉及领域较为广泛、获取困难的研究现状,提出了一种基于多特征融合的玉米前期图像旱情识别方法。以正常和特旱两种情况的玉米植株图像为样本,采用经典Kmeans算法对玉米植株图像提取感兴趣区域;进而提取分割后的玉米植株图像,包括颜色、奇异值分解(SVD)、纹理等共计20维特征;采用遗传算法对20维特征选择有效特征子集;最后针对有效特征子集建立了基于最小二乘支持向量机的判别模型,获取了玉米植株图像的旱情信息。将单个特征(颜色、SVD、纹理)直接融合之后的特征以及利用主成分分析法的特征选择作为对比实验,平均识别正确率分别为0.9503、0.9627、0.9771、0.9460、0.9745,而采用遗传算法进行特征选择后,最终寻到最优解为9维特征,平均识别正确率为0.9903。结果表明,运用图像处理技术可以对旱情进行识别,取得了较好的效果,为农业旱情的识别提供了新思路。
[Abstract]:In order to realize the drought analysis of maize plant, the current research status of agricultural drought index is more extensive and difficult to obtain. In this paper, a multi-feature fusion based drought recognition method for pre-maize images is proposed. The maize plant images in both normal and special drought conditions are taken as samples. The classical Kmeans algorithm is used to extract the region of interest from maize plant image. Then the segmented corn plant images were extracted, including color, singular value decomposition (SVD), texture and so on. Genetic algorithm is used to select an effective feature subset for 20 dimensional features. Finally, a discriminant model based on least squares support vector machine (LS-SVM) is established for the effective feature subset, and the drought information of maize plant image is obtained. Texture) after direct fusion and the feature selection of principal component analysis as a comparison experiment, the average recognition accuracy was 0.9503 / 0.9627 / 0.9771, respectively. The result shows that the optimal solution is 9 dimensional feature, and the average recognition accuracy is 0.9903. The image processing technology can be used to identify the drought, and a better result is obtained, which provides a new way of thinking for the recognition of agricultural drought.
【作者单位】：天津大学电气自动化与信息工程学院天津市过程检测与控制重点实验室;
【分类号】：TP391.41
【正文快照】： 1引言目前环境问题迫在眉睫,其中频发的重大干旱灾难尤为突出[1]。干旱本身作为一种复杂的现象难以直接观测其发生时间、发展过程以及影响范围。国内外对旱情的监测手段,从旱情统计上报到测定土壤含水量,发展到如今的遥感图像技术,利用多时相、高光谱遥感数据反映大面积的图

【相似文献】