ABSTRACT
The timing, convenient and reliable method of diagnosing and monitoring crop nutrition is the foundation of scientific fertilization management. However, this expectation cannot be fulfilled by traditional methods, which always need excessively work on sampling, detection and analysis and even exhibit lagging timing. In the present study, stable images for potassium-stressed leaf were acquired using stationary scanning, and object-oriented segmentation technique was adopted to produce image objects. Afterwards, nearest neighbor classifier integrated the spectral, shape and topologic information of image objects to precisely identify characteristics of potassium-stressed features. Diagnosing with image, the 3rd expanded leaves are superior to the 1st expanded leaves. In order to assess the result, 250 random samples and an error matrix were applied to undertake the accuracy assessment of identification. The results showed that the overall accuracy and kappa coefficient was 96.00% and 0.9453 respectively. The study offered an information extraction method for quantitative diagnosis of rice under potassium stress.
Subject(s)
Image Interpretation, Computer-Assisted , Oryza/chemistry , Plant Leaves/chemistry , Potassium/chemistry , Spectrum Analysis/methods , Stress, PhysiologicalABSTRACT
The real-time, effective and reliable method of identifying crop is the foundation of scientific management for crop in the precision agriculture. It is also one of the key techniques for the precision agriculture. However, this expectation cannot be fulfilled by the traditional pixel-based information extraction method with respect to complicated image processing and accurate objective identification. In the present study, visible-near infrared image of cotton was acquired using high-resolution sensor. Object-oriented segmentation technique was performed on the image to produce image objects and spatial/spectral features of cotton. Afterwards, nearest neighbor classifier integrated the spectral, shape and topologic information of image objects to precisely identify cotton according to various features. Finally, 300 random samples and an error matrix were applied to undertake the accuracy assessment of identification. Although errors and confusion exist, this method shows satisfying results with an overall accuracy of 96.33% and a KAPPA coefficient of 0.926 7, which can meet the demand of automatic management and decision-making in precision agriculture.