Abstract: The uniformity of gravel-soil blending is very important to the safety of gravel-soil core wall dams. Digital image processing methods are widely used in the field of mixture recognition, but over-segmentation or under-segmentation often occurs in processing images with adhesion characteristics, resulting in a low accuracy of evaluation. Aimed at these problems, this paper develops a new method for evaluating the uniformity of gravel-soil blending using an improved Harris-Susan algorithm. First, this algorithm is used to segment under-segmented regions and repair over-segmented regions so as to improve the accuracy of binary images. Then, the image’s uniformity of single quantile is evaluated using the concept of gravel neighborhood. Finally, particle distribution sampling is used to construct an evaluation index of spatial uniformity of gravel soil. We have applied this method to the construction of a gravel-soil high-core rockfill dam on the Dadu River at Shuangjiangkou, and achieved a relative statistical accuracy of 0.935, proving its significant improvement over the Otsu algorithm (0.353) or watershed algorithm (-3.101). This study shows that its results agree well with those of the screening method; it can effectively save manpower, material resources and time costs and it is applicable to engineering evaluation with high efficiency.

Key words: gravel-soil material, blending quality, digital image processing, Harris-Susan algorithm, evaluation of spatial uniformity