Detecting tree and wire entanglements with deep learning

Autor(es)

Oliveira, Artur Andre ; Buckeridge, Marcos S. ; Hirata Jr, Roberto

• Tipo de documento: Artigo Científico
• Fonte: TREES-STRUCTURE AND FUNCTION; v. N/A, p. 13-pg., 2022-05-08.

Resumo

Power and communication line corridors are usually mixed with urban trees, and this mixing can be the source of multiple issues like fires and communication failures. Nevertheless, urban trees are a valuable resource to the city as they dissipate heat island effects, reduce air pollution and increase general health perception. This work proposes a deep learning approach to detect trees entangled to power and communication lines using street-level imagery and perform quick quantitative and qualitative analyses based on the Grad-CAM++ method. Testing the method was performed using 1001 images from urban trees from the cities of Sao Paulo and Porto Alegre (both in Brazil). We found an overall accuracy of 74.6% (73.6% for Sao Paulo and 75.6% for Porto Alegre), suggesting that the methodology could be suitable in the future for city management to avoid risks of accidents due to contact between trees and electrical wiring. This text describes the method, a new data set of urban images, the experimental setup design and tests, and some possible future improvements. (AU)

• Processo FAPESP: 14/50937-1 - INCT 2014: da Internet do Futuro
• Beneficiário: Fabio Kon
• Modalidade de apoio: Auxílio à Pesquisa - Temático
• Processo FAPESP: 18/10767-0 - Investigar e analisar a cidade - INACITY
• Beneficiário: Artur André Almeida de Macedo Oliveira
• Modalidade de apoio: Bolsas no Brasil - Programa Capacitação - Treinamento Técnico
• Processo FAPESP: 15/24485-9 - Internet do futuro aplicada a cidades inteligentes
• Beneficiário: Fabio Kon
• Modalidade de apoio: Auxílio à Pesquisa - Temático
• Processo FAPESP: 15/22308-2 - Representações intermediárias em Ciência Computacional para descoberta de conhecimento
• Beneficiário: Roberto Marcondes Cesar Junior
• Modalidade de apoio: Auxílio à Pesquisa - Temático