Solving visual pollution with deep learning: A new nexus in environmental management.

Visual pollution is a relatively new concern amidst the existing plethora of mainstream environmental pollution, recommending the necessity for research to conceptualize, formalize, quantify and assess it from different dimensions. The purpose of this study is to create a new field of automated visual pollutant classification, harnessing the technological prowess of the 21st century for applications in environmental management. From the wide range of visual pollutants, four categories have been considered viz. (i) billboards and signage, (ii) telephone and communication wires, (iii) network and communication towers and (iv) street litter. The deep learning model used in this study simulates the human learning experience in the context of image recognition for visual pollutant classification by training and testing a convolutional neural network with several layers of artificial neurons. Data augmentation using image processing techniques and a train-test split ratio of 80:20 have been used. Training accuracy of 95% and validation accuracy of 85% have been achieved by the deep learning model. The results indicate that the upper limit of accuracy i.e. the asymptote, depends on the dataset size for this type of task. This study has several applications in environmental management. For example, the deployment of the trained model for processing of video/live footage from smartphone applications, closed-circuit television and drones/unmanned aerial vehicles can be applied for both the removal and management of visual pollutants in the natural and built environment. Furthermore, generating the 'visual pollution score/index' of urban regions such as towns and cities will create a new 'metric/indicator' in the field of urban environmental management.