Abstract: The ash content of flotation tailings is an important index of flotation products. In view of the fact that off-line detection is often used to measure the ash content of flotation tailings in coal preparation plants,the on-line accurate measurement can not be realized.And the current flotation soft sense mostly only bases gray image, which results in poor accuracy and adaptability of the soft-sensing model.This paper proposes a soft-sensing method of flotation tailings based on color image processing method, and establishes a soft-sensing model of flotation tailings ash based on least squares support vector machine (LSSVM). The model takes the color features of different color spaces, gray mean and concentration characteristics as input variables, and the ash content of tailings as output variables. Particle swarm optimization (PSO) algorithm is used to optimize the parameters of LSSVM model. The results show that the soft-sensing model of tailings ash can be used to predict the ash content of flotation tailings on-line. The accuracy of tailings image analysis can be improved by introducing color features of flotation tailings images, and the prediction accuracy can reach 96.89%. According to the research results of this experiment, the ash content of tailings can be well measured after its field application in Liuwan Coal preparation Plant.