Aiming at the problem of accurately predicting the ash content of tailings in the flotation process to achieve automatic control, a prediction method for the ash content of flotation tailings based on near-infrared image analysis was proposed based on near-infrared spectroscopy and image processing technology. A hybrid intelligent detection model for the ash content of flotation tailings was constructed by combining polynomial regression (PR) for preliminary prediction and random forest (RF) for compensation prediction. Twelve feature data including gray level and texture were extracted from the gray histogram and gray-level co-occurrence matrix of tailings images. After feature selection, a linear prediction model was established using PR, with a root mean square error (RMSE) of 2.752 and a coefficient of determination (R2) of 0.977. To improve the prediction effect, an RF compensation model was introduced on the basis of PR. The selected gray level and texture feature data were used as input, and the difference between the preliminary prediction value and the actual value was used as output. Finally, the preliminary prediction value and the compensation prediction value were added to obtain the ash content of flotation tailings, and a prediction model for the ash content of flotation tailings based on RF-PR hybrid was established. This model has high accuracy: in the low ash content range (11.1%-21.3%) and high ash content range (68.9%-76.2%), the difference between the preliminary prediction value and the offline laboratory value of tailings is reduced; in the medium ash content range (21.3%-68.9%), the mean absolute error (MAE) of the preliminary prediction value is reduced by 0.07. The results show that the RF-PR model has higher accuracy than the PR model and the RF model, and can meet the requirements of ash content detection of flotation tailings.