Inteligência artificial e sensoriamento remoto com o uso de VANTs: Uma aplicação na detecção automática de resíduos sólidos urbanos

• DOI: https://doi.org/10.22533/at.ed.7182427126

• Palavras-chave: Sensoriamento remoto, VANTs, Inteligência Artificial, Detecção automática de Resíduos sólidos

• Keywords: Remote sensing, UAVs, Automatic Waste Detection, Artificial Intelligence

• Abstract: According to the Institute of Applied Economic Research (IPEA), approximately 80 million tons of solid urban waste were generated in Brazil in 2018, a significant portion of which did not receive proper disposal and was irregularly discarded in urban or peripheral areas, slopes, and hard-to-reach areas. This problem is even more pronounced in regions where the state fails due to lack of presence and assistance, resulting in residents resorting to irregular waste disposal in hilly and hard-to-reach regions. In fact, even in more developed areas, the accumulation of such waste can lead to landslides and the proliferation of rats, cockroaches, flies, and other urban pests. Given the complexity of identifying irregular waste disposal sites in remote areas, either due to government omission or difficult access to these locations, this study aims to present unmanned aerial vehicles (UAVs) as an alternative for environmental monitoring. These devices have several advantages, such as mobility and real-time image acquisition from remote and hard-to-reach locations, with lower costs and no risks to the operator. However, due to the rugged terrain and vast territorial expanse of Rio de Janeiro, the volume of data collected by these devices would require significant effort for manual analysis and interpretation. To address this limitation, we believe that the images and videos collected in this survey should be automatically interpreted, georeferenced, and segmented using Artificial Intelligence. Thus, with a combined approach consisting of using an electronic device for remote sensing (UAV) equipped with high-precision cameras, sensors, and a global positioning system (GPS), along with computational resources for image processing and interpretation, this project aimed to demonstrate the feasibility of this technique in the early and effective identification of remote areas with irregular waste disposal. After a thorough remote sensing process carried out with the assistance of a UAV in the study areas of this project, the captured images were batch processed using digital image processing techniques and Deep Learning algorithms. After extensive training, these algorithms were able to automatically segregate and label images that contained evidence of solid waste disposal from those that did not. As a result of this automatic segregation, the technique demonstrated an 92% effectiveness in identifying specific types of waste, thus helping to automatically identify underserved areas that depend on government action to curb and mitigate irregular waste disposal and its environmental impact.