تحلیل حساسیت روش پردازش تصاویر در برآورد منحنی دانه بندی رسوبات سطحی بستررودخانه نسبت به اندازه سطح رسوبی

Some characteristics of rivers that are dependent upon particle-size distribution such as
riverbed grain-size (e.g., D50, D84 or D90), can be estimated using different methods which the
commonest is sieving analysis. Nowadays, recent innovations in image processing techniques
make it possible to determine the gradation curve through digital photographic methods.
Images are processed using a default or selective algorithm in most image processing
softwares including GIAS, ImageJ, and similar programs for Windows (e.g., Scion Image and
ImageTool) and other operating systems. Up to the present time, numerous studies were
conducted using image processing of sediment particles for a variety of purposes that mostly
is introducing the image processing as a replacement for traditional methods of particle size
determination (i.e., sieve analysis, Wolman pebble count, etc.) in order to save time, money,
energy and labor force. According to the importance of image processing and its application
in river engineering and drawing sediment gradation curve and also difficulties in field
surveys to take appropriate images of sediments at different points of the area, it seems
necessary to know the appropriate dimensions of sediment surface in order to accurate
processing of their particles. In our study, the accuracy of image processing technique (using
FHWA Hydraulic Toolbox software) to carry out an appropriate evaluation of grain-size of
surface layer sediments of Shalmanroud riverbed were evaluated by processing sedimentary
layers of two images with different size. In addition, usage possibility of the results obtained
using this method has evaluated for determining representative diameters of sediment
particles in some measuring methods for bed load.
After some visits from Shalmanroud River to select an appropriate study site, a 7.5-kilometer
length of the river was chosen and some photos of riverbed sediments and required samples
were taken for more examination. At 25 points along the river reach with uniform sediment,
some photos were taken from about one meter above riverbed surface at each point using a
10-megapixel Canon's PowerShot G12 digital camera and the available equipments. At each
point, two images were captured from above the sediments that one is related to surface
particles inside a 70cm×70cm wooden frame and the other is related to surface sediment
particles inside a 40cm×40cm wooden frame (which is fixed in the middle of the larger
frame). Then, the surface particles were gathered and stored in special storage bags for the
sieve analysis. At each point in the field, the UTM coordinates were measured using the
Garmin handheld GPS. The samples were delivered to the soil mechanics laboratory for sieve
analysis. In order to plot grain-size curve of sieve analysis, cumulative percentage retained on
each sieve was determined. Next, to achieve more accurate estimate of the grain-size
distribution using image processing technique, the FHWA Hydraulic Toolbox software was
used. Digital image processing consists of eight separate fundamental steps.
The results of image processing had been entered into Excel, and then particle-size
distribution curves were immediately obtained. Then, the values of representative diameters,
i.e., D16, D50, D75 and D84 were read from the particle-size distribution curves (laboratory and
software). After that, to examine the correlation and linear regression between the sieve result
and the result of image processing, correspondence analysis was run in Excel software and the
values of the correlation coefficients for D16, D50, D75 and D84 diameters were computed 0.93,
0.936, 0.905 and 0.824, respectively (obtained from processing of images using a
40cm×40cm wooden frame). In a similar way, those values were obtained for results of image
processing using the 70cm×70cm wooden frame as follows: 0.864, 0.876, 0.877 and 0.823,
shows a strong correlation between lab test results and image processing results. The results
revealed that the estimated diameters using the 40cm×40cm frame are closer to the
experimental values than what the 70cm×70cm frame estimated. It was also revealed that in
the case of using 40cm×40cm wooden frame, the estimation accuracy increases as the
diameters increase; whereas the converse results were obtained with the larger frame. It was
revealed that the error of estimate for finer particles was larger for particles smaller than
D50than that of particles larger than D50.
Image analysis using FHWA Hydraulic Toolbox software is sensitive to the size of the frame
so that in the case of using the smaller frame (40cm×40cm), the estimates of the particle size
are likely to be more accurate and show positive relationship with particle size while this
relationship turns negative when a larger frame is used. Regarding the wide range of particle
sizes (from very fine gravel to small pebble) on Shalmanroud Riverbed and also the higher
accuracy of image processing in estimating larger particles (in case of using the smaller
frame), it is highly recommended to use Meyer-Peter and Muller formula for computation of