摘要
【目的】针对在原始马铃薯病害图像上提取特征时计算量大、病害识别准确率低以及传统病害区域分割算法速度慢等问题,提出了一种新的基于关键特征点的病害感兴趣区域(ROI)快速检测与融合颜色和纹理特征的识别方法。【方法】对马铃薯病害图像作适当预处理后,首先提取ORB特征点,当其特征点数目小于给定阈值时提取SIFT特征点,再对所提特征点的坐标值按水平和垂直方向排序,并通过计算K个近邻点的均值来确定病害区域的坐标并提取ROI。然后融合病害ROI的HSV颜色直方图和UPLBP纹理直方图构成总特征向量。最后采用非线性SVM识别马铃薯病害。【结果】利用该方法对240幅马铃薯叶部、果实和茎部10种混合病害图像进行识别实验,结果表明,每幅病害图像ROI检测平均时间为0.013 s,平均识别正确率达95.83%,最高达100%,平均运行时间为0.083 s。【结论】基于ORB和SIFT关键特征点的病害ROI检测方法原理简单、易实现且实时性好。本文方法可实现对10类马铃薯病害的快速识别且准确率高,为其它农作物病害识别提供了参考价值。
【Objective】Aiming at the problem of extracting features from the original potato disease image, the problem of large amount of computation, low accuracy rate of disease recognition and slow speed of traditional disease area segmentation algorithm were discussed, and this paper proposed a new disease recognition method based on local feature points for rapid detection of Region of Interest(ROI) and fusion of color and texture features.【Method】After properly preprocessing the potato disease image, ORB feature points were firstly extracted, and SIFT feature points were extracted when the number of feature points was less than a given threshold. Then the coordinate values of the proposed feature points were sorted in the horizontal and vertical directions and calculated. The mean value of K nearest neighbor points was used to determine the coordinates of the disease region and ROI was extracted. Then, the histogram of the HSV color histogram and the UPLBP texture histogram of the ROI were merged to form a total feature vector. Finally, non-linear support vector machines were used to identify potato diseases.【Result】This method was used to identify the images of 10 mixed diseases of 240 potato leaves, fruits and stems. The results showed that the average time of ROI detection for each disease image was 0.013 s, and the average recognition accuracy rate was 95.83 %, up to 100 %. The average running time was 0.083 s.【Conclusion】The disease ROI detection method based on the key feature points of ORB and SIFT was simple, easy to realize and had good real-time performance. This method can be used for rapid identification of 10 kinds of potato diseases and high accuracy, which provided reference value for other crop disease identification.
作者
范振军
李小霞
FAN Zhen-jun;LI Xiao-xia(Robot Technology Used for Special Environment Key Laboratory of Sichuan Province,Mianyang Sichuan 621010,China;School of Information Engineering,Southwest University of Science and Technology,Mianyang Sichuan 621010,China)
出处
《西南农业学报》
CSCD
北大核心
2019年第3期544-550,共7页
Southwest China Journal of Agricultural Sciences
基金
国家自然科学基金项目(61771411)
西南科技大学研究生创新基金资助项目(17ycx123)
关键词
马铃薯病害
感兴趣区域
融合特征
支持向量机
识别
Potato disease
Region of interest ( ROI)
Fusion features
Support vector machine ( SVM)
Recognition