MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202641088062 A) filed by S. Hrushikesava Raju; and Koneru Lakshmaiah Education Foundation on July 19, 2026, for An Intelligent Hybrid Deep Learning Architecture For Air Quality Detection In Urbanized Environments.

Inventors include Dr. S. Hrushikesava Raju; Vempalli Sravani; Vogirala Nandini; Puliknatu Udaya Sree; K Karpagavalli; Thamodharan Arumugam; K. Selvam; and S Sherisha.

The application for the patent was published on July 24, 2026, under issue no. 30/2026.

Abstract: Diverse regions of any country such as Industrialization, Urbanization, and etc., where pollution assessment is required for initiating necessary steps to ensure good public health. There are traditional systems used in AQI classification that possess limitations that include inability to capture spatial temporal dependencies, limited feature representation, struggle with class imbalance issues, and absence of interpretability for predictions. To overcome these, require a hybrid approach that provides an enhanced explainable multimodal spatio-temporal framework for AQI classification. This hybrid model takes weather images and heterogeneous meteorological data as input, processes these images using CNN for extraction of local haze features and contextual data, whereas ViT for extraction of global atmospheric patterns. To capture pollution dynamics over time, use bidirectional temporal modeling using two way long short-term modelling (Bi-LSTM). Then, use deep neural network (DNN) for capturing structural environment components and adopt multi-modal fusion strategy that combines visual modalities and numerical modalities for improved robustness under the diverse conditions. The derived hybrid model finally classifies AQI into any of Good, Satisfactory, Moderate, Poor, Very Poor, and Severe, with later computation of focal loss to handle the class imbalances. In addition, incorporation of explainable AI techniques like Grad-CAM, SHAP, and LIME would make better predictions. Experimental evaluation states the hybrid classification achieves superior accuracy(99.1%), low error rate (0.9), ensures stability, and generalization across diverse regions, with real-time support and multi-horizon AQI forecasting.

Disclaimer: Curated by HT Syndication.