Automatic Vehicle Pollution Detection Using Feedback Based Iterative Deep Learning

Abstract

Air pollution is one of the major health hazards in modern times. Vehicle pollution is one of the major contributors to aerial contamination. Significant emphasis has been given by the researchers to identify the traffic pollutant sources. However, there is a need for a low-cost, automated solution for fast detection of polluting vehicles from the end of law enforcers. In this article, we have presented a novel deep learning based evaluation strategy that will identify the pollutant vehicle through on-road installed surveillance camera images. An enhanced image data set with notable variations has been prepared for training the models. Thereafter, a multi-model feedback process has been integrated. In contrast to the other deep learning approaches, the proposed framework has started with low labeled training data which is quite relevant in real-life scenarios. Subsequently, the size of the training samples has been increased using feedback. The evaluation of the framework has been performed with seven popular deep learning CNN models, i.e Inception-V3, MobileNet-V2, MobileNet-V3 (small), InceptionResNet-V2, VGG16, VGG19 and XceptionNet. Transfer learning has been exploited to distinguish the on-road pollutants and to provide proper surveillance in transport system. We have compared our method with recent state-of-the-art techniques. The results have demonstrated the superiority of the proposed framework in the road surveillance domain. To the best of our knowledge, this is the first attempt to apply a feedback based iterative deep learning model for vehicle pollution detection.