Abstract: Electrical engineering drawings are complex and diverse in elements, with varying quality, which affects the extraction of defect characteristics. Traditional methods for identifying defects in electrical engineering drawings mainly use fixed-point discrimination, but this method cannot identify highly similar defects and is prone to target overlap, leading to low detection accuracy. To address this, a defect detection technique based on generative adversarial networks for electrical engineering drawings is proposed. By extracting defect feature vectors from the drawings, high-similarity defect samples are generated through iteration. A two-class generative discrimination approach is used to calculate the weights of the defect samples, gradually resolving target overlap issues according to these weights, thus completing the defect detection of the drawings. Based on the generative adversarial network, the perceptual loss value for identifying defects within the coverage area is calculated. Using the perceptual loss target as guidance for defect reconstruction, the reconstructed defects are verified to achieve final detection and analysis. Experimental results show that the false detection rate of the proposed method is all within 1.5%, highlighting its high detection accuracy and superior performance.