FEW-SHOT IMAGE GENERATION METHOD FOR POWER DEFECT DETECTION

Electric power is an indispensable and important element in economic development and engineering construction, and the stable operation of the power system is of great significance. Once the power equipment has defects and failures, it will affect the safe and stable operation of the power system and have a significant impact on the country and society. At present, due to the scarcity of power defect data, most defect detection methods cannot effectively detect power defects accurately. Aiming at the problem of scarcity of defect data, this paper uses the method of few-shot image generation to propose an improved LoFGAN network, and designs a few-shot image generator based on context information, which improves the ability of the defect detection network to extract detailed features. The method based on LC-scatter regularized loss of divergence is introduced to optimize the training effect of image generation models on limited datasets. Experiments show that the few-shot image generation method proposed in this paper can generate real and diverse defect data for power scene defects. The improved LoFGAN surpasses the latest images such as FIGR and DAWSON Generate network.