Abstract: Large-capacity supercritical and ultra-supercritical circulating fluidized bed boilers are the mainstream of today's circulating fluidized bed combustion technology development, but the refractory wall of large-capacity circulating fluidized bed boilers lacks a fast and accurate crack measurement method with automated and intelligent inspection means. To address this problem, convolutional neural network and support vector machine are used to classify the images of cracked and uncracked wall to realize the intelligent identification of cracks,and at the same time, a soft measurement model of cracks is established to validate the accuracy and reliability of the digital image processing technology to measure the width and length of cracks, and then an example analysis is carried out to validate the images of cracks of the furnace wall inside the circulating fluidized bed boiler. The results show that under the condition of large amount of image data, the classification performance of convolutional neural network is better than that of support vector machine, and the accuracy rate reaches more than 90%; the measurement accuracy of digital image processing technology for cracks of different widths and shapes is different, and the digital image processing technology has the highest accuracy when the width of the crack is about 2 mm, and the digital image processing technology can recognize and measure the width of the crack efficiently. The actual crack width and length deviation is small, and the digital image processing technology can be adapted to the needs of use in the actual environment.