The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-3/W6
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W6, 467–471, 2019
https://doi.org/10.5194/isprs-archives-XLII-3-W6-467-2019
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W6, 467–471, 2019
https://doi.org/10.5194/isprs-archives-XLII-3-W6-467-2019

  26 Jul 2019

26 Jul 2019

RAPID IDENTIFICATION OF ABIOTIC STRESS (FROST) IN IN-FILED MAIZE CROP USING UAV REMOTE SENSING

J. Goswami1, V. Sharma1, B. U. Chaudhury2, and P. L. N. Raju1 J. Goswami et al.
  • 1North Eastern Space Applications Centre, Department of Space, Government of India, Meghalaya, India
  • 2ICAR Research Complex for NEH Region, Umiam, Ri-Bhoi district, Meghalaya – 79103, India

Keywords: Abiotic stress, Maize crop, Remote Sensing, UAV, Machine Learning, Classification

Abstract. Stress in the crop not only decreases the production but can also have devastating consequences for farmers whose life depends upon the healthy crops. In recent time (January 2018) a such abiotic stress event (hoar frost) was experienced at ICAR research complex experimental filed, Ri-Bhoi district of Meghalaya on standing Maize crop. Therefore, remote sensing (Multispectral UAV- Unmanned Aerial Vehicle) technology were used to detect the effect of frost on in-filed Maize crop. Two set of multispectral data (before frost and after frost) with four advanced machine learning techniques viz. Random Forest (RF), Random Committee (RC), Support Vector Machine (SVM) and Artificial Neural Network were employed for detection of stress free crop and stressed crop due to frost. Results revealed that all the four methods of classification could able to identify / detect stress-free vs. stressed crops at satisfactory level. However, among the classifiers RF achieved relatively higher overall accuracy (OA = 86.47%) with Kappa Indexanalysis (KIA = 0.80) and found very cost effective in context of computational cost (time complexity = 0.08 Seconds) to train the model. In addition, we have also recorded the area of each classes and found that after frost stress-free area (36.01% of all over filed) is decreased by 11% in comparison of before frost (25.036% of all over filed). Based on the results we can suggest that the RF ensemble classification method can be used for further other crop classification in order to estimate the yield, detect the condition, monitoring the health etc.