Abstract: To address the problems of large target scale variations,insufficient feature extraction,and low detection accuracy
in traditional remote sensing image target detection for mines,a method for mine remote sensing image target detection that
integrates Generative Adversarial Network (GAN) and Recurrent Neural Network (RNN) is proposed. This method first uses
an improved GAN to generate high-quality mine target samples to expand the training dataset. Then,a bidirectional RNN (Bi-
RNN) network is designed to extract temporal features,which are fused with spatial features extracted by a convolutional neural
network. Finally,the Faster R-CNN detection framework is adopted to achieve target detection. The performance of the algorithm
was verified on a test dataset containing typical mine targets such as open-pit mining areas,waste dumps and tailings
ponds. The results show that the average detection accuracy of this method reaches 92. 7%,which is 4. 3 percentage points
higher than that of Faster R-CNN. The recall rate for small targets has significantly improved from 85. 6% to 91. 2%. The detection
speed reaches 115 frames per second,providing a new technical means for mine safety monitoring and environmental assessment.

Key words: mine remote sensing image,target detection,generative adversarial network,recurrent neural network