FFA-GAN: A Generative Adversarial Network Based on Feature Fusion Attention for Intelligent Safety Monitoring

Abstract

With the rapid development of the national power grid, there is an increasing and strict demand for accurate intelligent management. However, the current detection algorithms have limited abilities under adverse conditions, especially in regions like Yunnan Province with complex terrain. To address this issue, we propose a method that utilizes infrared and visible images to make the images more informative, thereby improving the accuracy of the detection algorithm for electric power construction site safety. First, we design channel attention (CA) module and pixel attention (PA) module to focus on more important channels and resist thick haze pixels that focus on the thick haze pixels and more important channel information. Furthermore, we design a two-stage discriminator which imposes two restrictions on the fused results. Finally, we conduct a large number of comparison experiments with state-of-the-art methods, and the results show that our proposed fusion method achieves excellent performance in infrared and visible image fusion. This method has good prospects for application in the safety supervision of power construction sites and provides a line of defense for construction workers.