Detection of Diseased Tobacco Plants by Ground-based Remote Sensing

Abstract. In this paper, the real-time and reliable measurement on tobacco mosaic disease through field remote sensing was conducted using multi-spectral imaging. Multi-spectral images were acquired, using a Tetracam ADC multispectral camera equipped on a mobile platform at an altitude of 1.7 m, for 30 infected samples and 30 healthy samples at the canopy scale. First, the regions of interesting (ROI) of tobacco in the images were segmented using the GrabCut algorithm. Image texture feature including Gray-Level Co-occurrence Matrices (GLCM), and the vegetation index including NIR, R, G, NNI, NRI, NGI, N/R, N/G, R/G of the nine index were computed from region of interest (ROI). Then, Support vector machine (SVM) and Partial Least Squares (PLS) analysis were established for calibration to identify the healthy and diseased plants. The results showed the accuracy was undesired only using the single feature. The merging information was fed into SVM and PLS model to develop identifying models. Testing results of SVM model show that this method achieved the predicting accuracy of 85%, indicating field remote sensing based on multi-spectral image information could be applied for the detection of TMV at canopy scale.