A Spectral Similarity Measurement based on Piecewise Linear Representation

Spectral similarity measurement is a key technology for hyperspectral target classification. Although the commonly used similarity measurement methods based on high-dimensional space vector distance or angle have the advantage of intuitiveness, there are still limitations in practical applications, such as sensitivity to center wavelength drift. In response to the above issues, this paper proposes a spectral similarity measurement based on piecewise linear representation. This method evaluates similarity by analyzing the shape of spectral curves. The implementation process includes: firstly extracting key feature points on the spectral curve, then calculating the slope of the lines determined by adjacent feature points to form a sequence of slopes, and finally using the dynamic time warping algorithm to calculate the distance between different sequences to measure the similarity between spectra. This method has the following advantages: it does not require alignment of spectral bands and shows good robustness to noise and center wavelength drift. Experimental results on public datasets show that the target classification accuracy of this method is better than that of the traditional spectral angle similarity measurement. In addition, by introducing spatial feature information, this method can further improve classification accuracy and demonstrate promising application prospects.