Detection and Segmentation of Solar Farms in Satellite Imagery: A Study of Deep Neural Network Architectures
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Solar Farms, Detection, Satellite Images, Image Segmentation, Machine Learning, Deep Learning
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We develop a deep neural network that detects and segments solar farms in satellite imagery; its performance surpasses any previous result on a continent-spanning dataset reported in the literature.
Abstract: In line with global sustainability goals, such as the Paris Agreement, accurate mapping and monitoring of solar farms are critical for achieving net zero emissions by 2050. However, many solar installations remain undocumented, posing a challenge. This work introduces Solis-seg, a Deep Neural Network optimized for detecting solar farms in satellite imagery. Solis-seg achieves a mean Intersection over Union (IoU) of 96.26% on a European dataset, outperforming existing solutions.
The study leans heavily on advances in semantic segmentation and NAS for solar farm detection. Semantic segmentation has evolved through technologies like Fully Convolutional Network (FCN) and U-Net, which have shown strong performance on satellite imagery. In NAS, Differentiable Architecture Search (DARTS) and its variants like Auto-DeepLab (ADL) have become efficient ways to automate the creation of architectures. This study also challenges the prevailing method of using transfer learning from classification tasks for semantic segmentation, suggesting new avenues for research.
Thus, this work contributes to both the field of earth observation machine learning and the global transition to renewable energy by providing an efficient, scalable solution for tracking solar installations. We believe that our research offers valuable insights into the application of advanced machine learning techniques for solar farm detection and also encourages further exploration in earth observation and sustainability.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5182
Loading