•  
  •  
 

Abstract

The distribution of particle size in gravel-bed rivers affects the flow condition, sediment transport, and river morphology. However, conventional particle size analysis is laborious and time-consuming. To enhance the efficiency of particle size analysis, image-based methods have been widely explored, but complex particle textures increase the difficulty in identifying particle edges during image analysis. This study develops a filtering approach based on the rolling guidance filter and leverages the differences in grayscale gradients between particle edges and textures to design a guidance image tailored for the rolling guidance filter. This design enables the filter to preserve particle edges while smoothing particle textures. The filtering approach, referred to as the improved filter (IF), was used to evaluate the accuracy of image-based particle size analysis. Images for three types of particles, namely pebble, schist, and mixed particles, were used to assess the performance of five filters, including the anisotropic diffusion filter (ADF), bilateral filter (BF), guided filter (GF), median filter (MF), and the proposed improved filter (IF). The results demonstrated that the ADF, BF, GF, and MF could only enhance edges with large grayscale gradients and often misidentified edges with small grayscale gradients. The improved filter accurately processed edges of varying grayscale gradients and overcame the problem of weak edge recognition caused by the incorrect smoothing of particle images after filtering. The edge detection results obtained using the aforementioned filters revealed that the precision of detecting schist particle, pebble, and mixed particle edges ranged from 51.3% to 62.4%, 44.8% to 53.4%, and 43.3% to 48.7%, respectively. The IF demonstrated superior precision, exceeding the second-highest precision by 7.7%, 6.5%, and 4.9% for schist, pebble, and mixed particles, respectively. These results clearly indicate that the improved filter achieved the highest accuracy in edge detection for the different particle types because it effectively distinguished actual edges from the particles with weak surface textures as well as complex textures that in contact with the edges. Hence, the improved filter was able to enhance the accuracy of image-based particle size analysis.

Download
COinS