Combining spectral and texture features in hyperspectral image analysis for plant monitoring

Abstract

A texture enhanced spectral analysis framework is proposed for classifying hyperspectral images of plants of different conditions. Differentiating different plant conditions is important to precision agriculture as it helps detect diseases and stresses and optimise growth control. Advanced machine learning techniques are used to identify distinctive features in the spectral domain of hyperspectral images. In addition, texture properties are explored in the sub-band images. The framework integrates these two levels of properties at both feature extraction and classifying decision stages. The main crux of the work lies in the use of the significant spectral and texture features and a decision fusion mechanism to enhance the image properties, thus improving classification accuracy. Two hyperspectral datasets, originated from proximal hyperspectral systems, were used in the evaluation and significant improvements in classification accuracy achieved.