Particle size detection is a necessary condition for the realization of the blast furnace slag adaptive control system. The particle size has an important impact on energy recovery. A machine vision-based solution is proposed to solve the problem of real-time detection of blast furnace slag particle size. The accurate segmentation of the bonded particles puts forward higher requirements for image denoising. A denoising method based on adaptive median filtering and wavelet transform is proposed to smooth the image, while filtering out most of the noise. Protects the details and edges of the image. In order to verify its denoising effect, the MATLAB simulation is used to improve the segmentation effect of the watershed algorithm as the evaluation standard, which indicates that this method meets the test requirements for the denoising effect of blast furnace slag image, and provides technical support for accurate segmentation and extraction of particle size.