A feature-supervised generative adversarial network for environmental monitoring during hazy days

Ke Wang,Siyuan Zhang,Junlan Chen,Fan Ren,Lei Xiao

DOI: https://doi.org/10.1016/j.scitotenv.2020.141445

2020-12-01

Abstract:<p>The adverse haze weather condition has brought considerable difficulties in vision-based environmental applications. While, until now, most of the existing environmental monitoring studies are under ordinary conditions, and the studies of complex haze weather conditions have been ignored. Thence, this paper proposes a feature-supervised learning network based on generative adversarial networks (GAN) for environmental monitoring during hazy days. Its main idea is to train the model under the supervision of feature maps from the ground truth. Four key technical contributions are made in the paper. First, pairs of hazy and clean images are used as inputs to supervise the encoding process and obtain high-quality feature maps. Second, the basic GAN formulation is modified by introducing perception loss, style loss, and feature regularization loss to generate better results. Third, multi-scale images are applied as the input to enhance the performance of discriminator. Finally, a hazy remote sensing dataset is created for testing our dehazing method and environmental detection. Extensive experimental results show that the proposed method has achieved better performance than current state-of-the-art methods on both synthetic datasets and real-world remote sensing images.</p>

environmental sciences

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address the issue of image dehazing in environmental monitoring under hazy weather conditions. Specifically, the paper proposes a feature-supervised learning network based on Generative Adversarial Networks (GAN) for environmental monitoring during hazy days. The main objectives are:

1. **Improve Image Quality**: By training the model to obtain high-quality feature maps, thereby generating clear, haze-free images.
2. **Enhance Existing Methods**: Most existing environmental monitoring research is conducted under normal conditions, with fewer studies focusing on complex hazy weather conditions. Therefore, this paper attempts to fill this gap.
3. **Multi-Scale Discriminator**: Introduces a multi-scale discriminator to enhance model performance and proposes several loss functions to constrain the training process.
4. **Create Dataset**: Created a remote sensing dataset containing synthetic and real hazy images to test the dehazing methods and the effectiveness of environmental detection.

Through these technical means, the method proposed in this paper achieves better performance on both synthetic datasets and real-world remote sensing images compared to the current state-of-the-art methods.