Abstract: Based on the principle of light scattering, measuring dust concentration can only be done manually at fixed times and locations, with poor real-time performance. It can only detect dust mass concentration and cannot provide a range of particle size distribution. At present, research on dust particle detection based on image analysis mainly focuses on unilateral research on dust mass concentration or particle size distribution. It cannot achieve simultaneous detection of dust mass concentration and particle size distribution range. In order to solve the above problems, a method for detecting dust particles in excavation working face based on image analysis is proposed. It explores the relationship between image features and dust mass concentration and particle size distribution. By using a dust sample collection and image acquisition device, dust particle images are collected and the dust mass concentration at the time of image acquisition is obtained. An image processing algorithm for dust samples, is developed to extract parameters related to grayscale features, texture features, and geometric features of the image. The correlation analysis between the extracted image features and the measured dust mass concentration is performed, and the image features with high correlation is selected as parameters to establish a regression mathematical model. The method extracts the number of pixels in the dust particle object. Combining with the conversion coefficient, the method calculates the size and distribution range of the dust particle based on the geometric equivalent area diameter. The experimental results show that the average relative error between the measured dust mass concentration and the calculated values of the established image feature multiple nonlinear regression model mathematical model is 12.37%. The maximum relative error between the standard measured particle size and the geometric equivalent area size obtained from the particle size distribution is 8.63%, and the average relative error is 6.37%. This verifies the accuracy of the image feature based dust mass concentration regression mathematical model and the geometric equivalent area diameter distribution mathematical model.