Abstract: The existence of the abnormal images will affect the calculation accuracy while evaluating the indexes such as particle size distribution, roundness, and sphericity in oilfield hydraulic fracturing using image analysis methods. In order to solve this problem, this paper proposed different algorithms to remove the abnormal quarts sand images from three different scale datasets: 20/40 mesh, 40/70 mesh and 70/140 mesh. Firstly, research on the Laplacian variance-based blurred image detection algorithm was carried out, and the traditional Fast Fourier Transform was compared. Secondly, the convex hull-based detection algorithm was proposed to calculate the convex defects of quarts sand particles and the distance from the convex defects to the convex hull. It was used to determine whether the particles are collided. Based on the analysis of experiments, the accuracy of the proposed Laplacian variance method is 51.67% higher than Fast Fourier Transform algorithm. For the 20/40 mesh and 40/70 mesh datasets with larger quartz sand particles, the accuracy of the collided particle image detection based on convex hull algorithm proposed in this paper can reach to 88.38% and 82.63%, respectively. For the 70/140 mesh dataset with relatively smaller particles and uneven size distribution, a hierarchical strategy was proposed, which improves the accuracy of detection by 10.81 percentage points. This study shows that image analysis methods can effectively detect blurred and collided particles in quartz sand images to improve the quality of the dataset, and also provide support for the accuracy of quartz sand proppant particle size and shape calculation.