Abstract: The current situation of coal-fired boiler operation under wide load and co-firing of biomass has exacerbated the risk of slagging on the heating surfaces of the boiler. An online identification method for the slagging state of the boiler heating surfaces based on machine vision was proposed, which differed from the extensive monitoring model relying on "manual observation." First, the acquired raw slagging images were pre-processed to reduce interference and enhance the probability of detecting key features through image denoising, along with image enhancement and transformation to strengthen the image dataset. Next, a comparative analysis was conducted on the slagging identification effects of three convolutional neural network algorithms (AlexNet, VGG16, and ResNet34) and three optimizers (SGDM, Adam, and RMSProp) to determine the optimal combination of network model and optimizer. Finally, a digital online monitoring system for the slagging state of the boiler heating surfaces was designed, and applied to a certain in-service supercritical 600 MW coal-fired boiler. The results show that the selected ResNet34 convolutional neural network algorithm and Adam optimizer can achieve a slagging state identification accuracy of 100%. The proposed method enables online monitoring and identification of the slagging state of boiler heating surfaces, which is of great significance for ensuring the heat transfer efficiency and operational safety of the boiler.