In order to improve the retrieval performance of frozen power line images, a deep adversarial hashing method based on auto-encoder was proposed. First, a new code discriminator was added on top of the existing generator and discriminator to encourage the generated image samples to better represent the real data distribution. Second, a hash coding network was built to learn to generate a compact binary hash code. On the basis of the WGAN-GP loss, we introduced new cross-entropy loss and quantization loss functions based on the long-tail Cauchy distribution to optimize the Hamming space retrieval performance. The experimental results show that the deep adversarial hashing method can solve the problems of mode collapse and image blurring through the code discriminator and Cauchy loss functions, and the image retrieval performance is significantly improved with respect to other methods.