Petrochemical storage tanks are susceptible to long-term erosion from internal and external environments during service, which poses significant risks to ecological safety, national economy, and residential security. Current machine vision technologies including deep learning and traditional image processing show limitations in corrosion detection and identification, with recognition results generally lacking objective quantitative standards. Aiming at the above problems, a rust detection method for petrochemical storage tanks based on super-pixel segmentation was proposed. By building a lightweight convolutional neural networks (LCNN) model, the image was initially segmented and super-pixels were generated using simple linear iterative clustering (SLIC) algorithm, and then the fusion of fully connected conditional random field (CRF) algorithm with texture kernel was used to accurately segment the rust and determine the degree of rust using red-green-blue (RGB) color space. The experiments show that compared with the detection methods such as deep learning and traditional image processing, the petrochemical storage tank rust detection method based on super-pixel segmentation can improve the detection efficiency and accuracy of petrochemical storage tank corrosion to a certain extent.