Grain‐size analysis is arguably at the base of sediment analysis, but shape analysis is often reduced to empirical equations from a few parameters and micro‐variability within a single grain‐size class from one grain to another is a methodology in its infancy. In this study, we investigate an image analysis technique to automatically analyze the particle shape and roughness using gravels from pyroclastic flow and debris‐flow and water‐borne transport deposits at Unzen Volcano. The purpose of this study is to investigate the effective shape coefficients and to evaluate the image analysis method by classifying the sediment using the shape coefficients and contour signals, in order to define debris‐flow deposits from pyroclastic‐flow material. The results of the Principal Component analysis suggested that the shape coefficients could be a first indicator to classify sediments with different depositional processes. From this dataset, a logistic regression analysis, which predicts the location of sediment sampling (Gully or Wall) has also shown that the shape coefficient AR (Aspect Ratio) provides a differentiation based on the sediment transport process, but it is the wavelet analysis using the particle contour signal that expressed the clearer distinction between the different samples. It showed that the amplitude is larger for the walls than for the gully in the high frequency band. This indicates that the surface of the sample collected from the walls has fine irregularities, compared to those that travelled in the channel. It was thus found that it is possible to differentiate otherwise homogeneous sediments coming from the same source, just based on their transport process, using shape and roughness information, even if the grain‐size does not provide a clear distinction between the samples. The entire procedure from image processing to the calculation of shape coefficients was performed in Python, making the analysis inexpensive, simple, and flexible.