Abstract

We study the pattern spectra in context of soil structure analysis. Good soil structure is vital for sustainable crop growth. Accurate and fast measuring methods can contribute greatly to soil management decisions. However, the current in-field approaches contain a degree of subjectivity, while obtaining quantifiable results through laboratory techniques typically involves sieving the soil which is labour- and time-intensive. We aim to replace this physical sieving process through image analysis, and investigate the effectiveness of pattern spectra to capture the size distribution of the soil aggregates. We calculate the pattern spectra from partitioning hierarchies in addition to the traditional max-tree. The study is posed as an image retrieval problem, and confirms the ability of pattern spectra and suitability of different partitioning trees to re-identify soil samples in different arrangements and scales.