Gabor wavelet image analysis for soil texture classification

Soil texture is an important physical property of soil that affects many agricultural activities. It describes soil composition in terms of the relative proportion of three typical sized particles, i.e., clay, silt and sand. Traditional soil texture analysis methods involve inefficient physical and chemical processing procedures. To improve the efficiency for the analysis, previously we proposed a wavelet frame based image analysis system that related textural patterns observed at soil surface to the particle compositions. The system was capable of differentiating between 33 soil samples in terms of three categories with a 91% success rate. However, it required image acquisition under two camera settings. In this paper, we further our investigation with an improved image analysis approach, in which Gabor wavelets are utilized to generate textural features. Experiments showed that a combination of analysis results from two groups of Gabor wavelets yielded a 91% classification accuracy. Although the accuracy remained unchanged, the Gabor wavelet based system provided improved efficiency and flexibility over the previous system in that it needs only one set of images acquired under a fixed camera setting. Moreover, an improved consistency between individual classification votes was observed with the new system, indicating a greater potential for a finer categorization of soil textures.