Roof-CNN: Convolutional Neural Network Model with Less Parameters for Semantic Segmentation of Buildings

Abstract

Today, solar energy has become an indispensable element in providing energy infrastructure. Estimating the solar energy potential of building roofs in residential areas is important for the effective use of this energy. Nowadays, thanks to the developments in artificial intelligence algorithms, these tasks are performed automatically by computers. In this study, a solution is proposed by using deep learning architectures, which are the most advanced artificial intelligence algorithms. Convolutional Neural Network model called Roof-KSA with less parameters was proposed for semantic segmentation of building roofs in this research. A total of 3400 satellite images in 224×224×3 pixels size were used for semantic segmentation. Roof-CNN model has a total of 10 layers and 104,450 updated parameters. Within the scope of comparative analysis, Roof-CNN model has less parameters compared to U-Net models. In addition, Roof-KSA model stands out with 0.91404 global accuracy, 0.73092 mean accuracy, 0.65537 mean intersection over union, 0.84918 weighted intersection over union and 0.67244 mean BF score. As a result, it is seen that Roof-CNN model is more successful in accordance with obtained semantic segmentation results.

Keywords

• Solar Energy
• Deep Learning
• Semantic Segmentation
• Convolutional Neural Networks

Roof-KSA: Binaların Semantik Bölütlemesi İçin Az Parametreye Sahip Konvolüsyonel Sinir Ağı Modeli

Abstract

Günümüzde güneş enerjisi, enerji alt yapısının sağlanmasında vazgeçilmez bir unsur haline gelmiştir. Yerleşim yerlerinde bulunan bina çatılarının güneş enerjisi potansiyelinin tahmin edilebilmesi bu enerjinin etkin kullanılabilmesi için önemlidir. Günümüzde yapay zeka algoritmalarındaki gelişmeler sayesinde bu gibi işler bilgisayarlar tarafından otomatik olarak gerçekleştirilmektedir. Bu çalışmada ise yapay zeka algoritmalarının en gelişmişi olan derin öğrenme mimarilerinden faydalanarak bir çözüm önerilmektedir. Bu çalışmada bina çatılarının semantik bölütlenmesi için az parametreye sahip Roof-KSA adı verilen Konvolüsyonel Sinir Ağı modeli önerilmiştir. Semantik bölütleme işlemi için toplamda 3400 adet 224×224×3 piksel boyutlarında uydu görüntülerinden yararlanılmıştır. Roof-KSA modeli toplam 10 katmana ve 104,450 adet güncellenebilen parametreye sahiptir. Karşılaştırmalı analiz kapsamında Roof-KSA modeli kullanılan U-Net modellerine göre oldukça az parametreye sahiptir. Ayrıca Roof-KSA modeli 0.91404 küresel doğruluk oranı, 0.73092 ortalama doğruluk oranı, 0.65537 ortalama eşleşmiş bölge oranı, 0.84918 ağırlıklandırılmış eşleşmiş bölge oranı ve 0.67244 ortalama BF skoru ile ön plana çıkmaktadır. Elde edilen semantik bölütleme sonuçları dikkate alındığında Roof-KSA modelinin oldukça başarılı olduğu görülmektedir.

Keywords

• Güneş Enerjisi
• Derin Öğrenme
• Semantik Bölütleme
• Konvolüsyonel Sinir Ağları

References

1. Harrouz, A., Colak, I., Kayisli, K. (2019). Energy Modeling Output of Wind System based on Wind Speed. In 2019 8th International Conference on Renewable Energy Research and Applications (ICRERA) (pp. 63-68). IEEE.
2. Rehman, A., Rauf, A., Ahmad, M., Chandio, A. A., Deyuan, Z. (2019). The effect of carbon dioxide emission and the consumption of electrical energy, fossil fuel energy, and renewable energy, on economic performance: evidence from Pakistan. Environmental Science and Pollution Research, 26(21), 21760-21773.
3. Johlas, H., Witherby, S., Doyle, J. R. (2020). Storage requirements for high grid penetration of wind and solar power for the MISO region of North America: A case study. Renewable Energy, 146, 1315-1324.
4. Robles, E., Haro-Larrode, M., Santos-Mugica, M., Etxegarai, A., Tedeschi, E. (2019). Comparative analysis of European grid codes relevant to offshore renewable energy installations. Renewable and Sustainable Energy Reviews, 102, 171-185.
5. Akcan, E., Kuncan, M., Minaz, M. R. (2020). PVsyst Yazılımı İle 30 Kw Şebekeye Bağlı Fotovoltaik Sistemin Modellenmesi ve Simülasyonu. Avrupa Bilim ve Teknoloji Dergisi, (18), 248-261.
6. Cui, Y., Yao, H., Zhang, J., Zhang, T., Wang, Y., Hong, L., Wei, Z. (2019). Over 16% efficiency organic photovoltaic cells enabled by a chlorinated acceptor with increased open-circuit voltages. Nature communications, 10(1), 1-8.
7. Khasraw Bani, R., Jalal, S. J. (2019). Impact of shadow distribution on optimizing insolation exposure of roofs according to harness or transfer of solar energy in Sulaimani city, Iraq. Renewable energy, 136, 452-462.
8. Eia, U. S. (2017). The international energy outlook 2016.

9. Wu, G., Yang, Q., Fang, H., Zhang, Y., Zheng, H., Zhu, Z., Feng, C. (2019). Photothermal/day lighting performance analysis of a multifunctional solid compound parabolic concentrator for an active solar greenhouse roof. Solar Energy, 180, 92-103.
10. Victoria, M., Andresen, G. B. (2019). Using validated reanalysis data to investigate the impact of the PV system configurations at high penetration levels in European countries. Progress in Photovoltaics: Research and Applications, 27(7), 576-592.
11. Ghimire, S., Deo, R. C., Downs, N. J., Raj, N. (2019). Global solar radiation prediction by ANN integrated with European Centre for medium range weather forecast fields in solar rich cities of Queensland Australia. Journal of cleaner production, 216, 288-310.
12. Chen, C., Duan, S., Cai, T., Liu, B. (2011). Online 24-h solar power forecasting based on weather type classification using artificial neural network. Solar energy, 85(11), 2856-2870.
13. Shen, S., Yang, H., Yao, X., Li, J., Xu, G., Sheng, M. (2020). Ship Type Classification by Convolutional Neural Networks with Auditory-Like Mechanisms. Sensors, 20(1), 253.
14. Zhou, D. X. (2020). Universality of deep convolutional neural networks. Applied and computational harmonic analysis, 48(2), 787-794.
15. Mnih, V. (2013). Machine learning for aerial image labeling. University of Toronto (Canada).
16. Hassabis, D., Kumaran, D., Summerfield, C., Botvinick, M. (2017). Neuroscience-inspired artificial intelligence. Neuron, 95(2), 245-258.
17. Nagy, G., Anderson, J. (2016, May). Active recruitment mechanisms for heterogeneous robot teams in dangerous environments. In Canadian Conference on Artificial Intelligence (pp. 276-281).
18. Marr, B. (2018). Is Artificial Intelligence dangerous? 6 AI risks everyone should know about. Forbes.
19. Krizhevsky, A., Sutskever, I., Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105).
20. Wei, Y., Xiao, H., Shi, H., Jie, Z., Feng, J., Huang, T. S. (2018). Revisiting dilated convolution: A simple approach for weakly-and semi-supervised semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 7268-7277).
21. Hu, W., Huang, Y., Wei, L., Zhang, F., Li, H. (2015). Deep convolutional neural networks for hyperspectral image classification. Journal of Sensors, 2015.