A Deep Learning Approach for Air Pollution Classification Using InceptionV3 with Transfer Learning

Abstract

Pollution classification and severity assessment are crucial tasks in environmental monitoring and public health management. This study proposes a deep learning-based approach for pollution classification and severity assessment utilizing publicly available air quality datasets from India and Nepal. The similarity in the climatic conditions in North-eastern part of India and Nepal made us to choose the dataset. The proposed methodology employs transfer learning with the InceptionV3 architecture that had been pre-trained on the ImageNet dataset. By fine-tuning pre-trained weights, the model effectively captures complex features inherent in pollution images. After fine-tuning the pre-trained model, regularization techniques such as dropout and L2 regularization is used to enhance performance and generalization. The model is trained on a comprehensive dataset comprising three classes: “Good”, “Moderate”, and “Severe”, representing different levels of pollution severity commonly observed in the Indian and Nepalese regions. The novelty of the model is that it offers real-time assessments of pollution severity levels through a user-friendly interface. Any user can use the webapp that has been created to upload the images and can get to know the quality of the environment. This allows the user to know the on-spot evaluation of the surrounding environment. Through experimentation and evaluation, the proposed approach demonstrates good results in pollution classification with an accuracy of 92.65%. The usage of Inception V3 along with the Transfer Learning has given us this accuracy which was ahead VGG and Inception V3 without Transfer learning. The model provides actionable insights by categorizing pollution severity levels based on the input image which is given as an input. If any input given by the user doesn’t come under the three classes that is “Good”, “Moderate” and “Severe”, the user will get the result as “Irrelevant data”.