Power Line Aerial Image Restoration under dverse Weather: Datasets and Baselines

Abstract

Power Line Autonomous Inspection (PLAI) plays a crucial role in the construction of smart grids due to its great advantages of low cost, high efficiency, and safe operation. PLAI is completed by accurately detecting the electrical components and defects in the aerial images captured by Unmanned Aerial Vehicles (UAVs). However, the visible quality of aerial images is inevitably degraded by adverse weather like haze, rain, or snow, which are found to drastically decrease the detection accuracy in our research. To circumvent this problem, we propose a new task of Power Line Aerial Image Restoration under Adverse Weather (PLAIR-AW), which aims to recover clean and high-quality images from degraded images with bad weather thus improving detection performance for PLAI. In this context, we are the first to release numerous corresponding datasets, namely, HazeCPLID, HazeTTPLA, HazeInsPLAD for power line aerial image dehazing, RainCPLID, RainTTPLA, RainInsPLAD for power line aerial image deraining, SnowCPLID, SnowInsPLAD for power line aerial image desnowing, which are synthesized upon the public power line aerial image datasets of CPLID, TTPLA, InsPLAD following the mathematical models. Meanwhile, we select numerous state-of-the-art methods from image restoration community as the baseline methods for PLAIR-AW. At last, we conduct large-scale empirical experiments to evaluate the performance of baseline methods on the proposed datasets. The proposed datasets and trained models are available at https://github.com/ntuhubin/PLAIR-AW.

Figures (7)

• Figure 1: (a) Average precision (AP) of the advanced real-time instance segmentation model on a popular dataset of TTPLA under Normal, Hazy, Rainy, and Snowy conditions, respectively. There is a substantial decline in adverse weather compared with the normal situation. (b) The visual comparison between normal and hazy conditions, suggests that missed and false detection exists in hazy cases. (c) The general solution framework is based on deep learning for Power Line Aerial Image Restoration under Adverse Weather (PLAIR-AW). The power line aerial images under adverse weather are input into the deep neural network, which outputs the restored images. Then, the objective function is established between the ground-truths and the restored images. The deep neural network is optimized with the objective function. After training, the deep neural network is deployed to do the PLAIR-AW test.
• Figure 2: (a) The exemplar images of the proposed HazeCPLID, HazeTTPLA, HazeInsPLAD, RainCPLID-L, RainTTPLA-H, RainTTPLA-L, RainTTPLA-H, RainInsPLAD-L, RainInsPLAD-H, SnowCPLID-S, SnowCPLID-M, SnowCPLID-H, SnowTTPLA-S, SnowCPLID-M, SnowCPLID-H datasets. (b) The information loss of each proposed dataset is compared with its clean counterpart, which is measured by the Peak Signal-to-Noise Ratio (PSNR). (c) The performance ranking of the baseline methods.
• Figure 3: Illustration of baseline methods for the new Power Line Aerial Image Restoration under Adverse Weather (PLAIR-AW) task. These baseline methods are broadly categorized into three families of ($\rm I$) Single adverse weather removal, ($\rm II$) Multiple adverse weather removal, and ($\rm III$) All-in-one adverse weather removal. In the first family, we illustrate the representative methods for each specific adverse weather removal, namely, (a) FFANet qin2020ffa, (b) AECR-Net wu2021contrastive, (c) Dehazeformer song2023vision for power line aerial image dehazing, (d) PReNet ren2019progressive, (e) DRSformer chen2023learning for power line aerial image deraining, (f) LMQFormer lin2023lmqformer for power line aerial image desnowing. The representative methods in the second family include (g) SwinIR liang2021swinir, (h) Uformer wang2022uformer, (i) Restormer zamir2022restormer (j) CAT chen2022cross, (k) Stoformer xiao2022stochastic, (l) ShuffleFormer xiao2023random, (m) CODE zhao2023comprehensive, (n) ART 2023ART, (o) GRL li2023efficient. The representative methods in the third family include (p) AirNet li2022all, (q) TransWeather valanarasu2022transweather, (r) PromptIR potlapalli2023promptir.
• Figure 4: Visual comparison results on power line aerial image dehazing task in single-one setting. Please zoom in on the figure for a better view.
• Figure 5: Visual comparison results on power line aerial image deraining task in single-one setting. Please zoom in the figure for a better view.