IDENTIFICATION OF BARLEY SCAB BASED ON MULTI-SPECTRAL IMAGING SENSOR
Site-specific variable pesticide application is one of the major precision crop production management operations. Traditional chemical methods can do the accurate barley scab identification; however they are time-consuming, high cost and requiring execution by professionals. Barley scab identification and classification by human eyes needs special crop protect knowledge, but it is also low efficiency. To obtain sufficient barley disease information is essential for achieving effective site-specific pesticide applications and crop management. A method for real-time and reliable detection of barley disease is developed in this paper. The infected barley and the healthy barley leaf image samples were collected by multi-spectral imaging sensor. The backgrounds of images were removed by using threshold segmentation algorithm in the near-infrared channel image. The barley awn was removed by using morphological opening operation which is a combination of dilation and erosion. Then twelve statistical characteristics including the mean values and variances of the gray values of the image components in RGB color space and HIS color space of images were captured from the multi-spectral image being pretreated. The statistical characteristics were pretreated by using standard normal variate (SNV) method, MSC (multiplicative scatter correction) method and Savitzky-Golay smoothing plus SNV method respectively. Partial least squares (PLS) analysis was applied as calibration method as well as a way to extract the principal components which could be used to represent the most useful information of original image data and compress the data dimensionality. The selected principal components were used as the input data matrix of least squares-support vector machine (LS-SVM) to develop LS-SVM identifying model. The calibration set was composed of 60 samples, the validation set 49 samples. The MSC pretreatment method is the best method compared to others and the first principal component was recommended as the input variables by the PLS analysis. The predicting accuracy of this LS-SVM model was 93.9％. The results indicated that the method of identifying barley scab based on multi-spectral images was feasible. And a new approach was proposed for detecting the plant diseases and pests infection. Thus, it is concluded that multi-spectral imaging technique is available for the detection of barley scab on the barley spike.
Barley scab Multi-spectral image Partial least squares analysis Least square-support vector machine
SUN Guangming YANG Kaisheng FENG Lei ZHANG Chuanqing WU Di HE Yong
College of Biosystems Engineering and Food Science,Zhejiang University,Hangzhou 310029,China College of Agriculture and Food Science,Zhejiang Forestry University,Hangzhou 311300,China
国际会议
北京
英文
1-7
2009-10-14（万方平台首次上网日期，不代表论文的发表时间）