When explosive dust concentrations reach a certain value near an ignition source, it creates an explosion hazard. Therefore, we propose an explosive dust detection and characteristic analysis method using dust particle imagery. First, the Fourier transform domain of the fractional derivative is used to define the image filtering framework. The Rudin – Osher – Fatemi (ROF) model, in a bounded variation imagery function space, is selected to obtain prior noise knowledge with noise variance. Then, the imagery texture region and the nontexture region are divided according to the statistical information of the image local variance. A modified differential evolution particle swarm optimization algorithm is then used to identify the complex dust particles and to determine and update the fitting parameter optimal values, which can separate the overlapping particle intersection points. The model and algorithm are compared and analysed experimentally. The influence of the dust particle parameters is then obtained. We thus demonstrate that the noise suppression and staircase effect are better for the dust images, and that the overlapping particles are effectively separated. The research results indicate the correctness and feasibility of the proposed model, which provides the theoretical and experimental basis for the design of dust explosion concentration intervals.