Extraction and Analysis of Farmland Objects in Hyperspectral Images

One of the objectives of precision agriculture is rapid location of fields information to reduce the investment and improve the environment. Based on near infrared (NIR) hyperspectral images, this paper outlines an approach using hyperspectral imaging technology, combining spectral analysis methods and supervised classification methods for the extraction and analysis of farmland objects. This paper selected 185 bands hyperspectral image data and used spectral angle mapper, binary encoding and spectral information divergence to extract farmland objects and assess classification accuracy based on the extraction results. The N-dimensional visualization analyzer is used to purify the reference pixels. The performance of the provided methods is shown for a series of images acquired under varying light, different fields and different times. The average classification accuracies are 85.3%, 88.0% and 93.9%. The results show that spectral information divergence outperforms in accuracy and extraction effect, and it will provide a reliable and rapid method in image processing.