Abstract: Oil production and transportation safety is essential for ensuring a stable energy supply. Using robots to monitor the condition of oil production equipment can effectively prevent accidents and reduce risks. This paper presents a method for detecting leakage states in oil pipelines. First, a feature enhancement module based on residual pyramid structure is designed to capture image details and improve the discrimination between leakage state features and complex backgrounds through its multi-scale analysis capability. Then, a model compression strategy based on information entropy metrics is proposed to improve detection efficiency under edge device deployment conditions. Experiments conducted on a real dataset from an oil field demonstrate the effectiveness of the proposed method. Compared to the baseline model, the model parameter compression rate increased to 56.43%, achieving a 44.29% reduction in floating-point operations and a 28.60% reduction in parameter count, while detection accuracy decreased by only 0.02%.