Journal of Image Processing and Artificial Intelligence

Ward-Level Urban Climate Risk Intelligence Using Google Earth Engine and Hybrid Deep Learning Forecasting

2026-09-21T11:06:54+00:00

Rapid urbanization is increasing urban climate risk by raising surface temperatures, reducing vegetation-based cooling, increasing electricity and water demand, and exposing more residents to heat-related stress. This paper presents a ward-level urban climate risk intelligence framework for Bengaluru, India. The framework integrates Google Earth Engine (GEE), Landsat thermal and spectral indicators, measured ward population density, built-up measurements, deep learning-based validation, and local large language model-assisted action reporting. Instead of producing only a city-level heat map, the proposed framework generates analytical outputs for municipal planning: ward-level risk surfaces, sector-specific risk scores, model validation metrics, intervention priority ranks, and climate action reports. The GEE export contains 20,412 valid ward-month observations from 243 wards across the seven-year period 2018-2024. Each monthly record contains Land Surface Temperature (LST), Surface Urban Heat Island (SUHI), Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Normalized Difference Built-Up Index (NDBI), population density, and built-up fraction. Five risk sectors are assessed: heatwave risk, power grid stress, water scarcity risk, urban fire risk, and public health risk. A feed-forward multilayer perceptron estimates present ward risk from current normalized indicators. In parallel, GRU, LSTM, and MLP models are trained on monthly sequences from 2018-2023 and compared for 2024 risk prediction, forming a hybrid architecture that combines transparent risk formulas with data-driven temporal validation. The present implementation identifies 17 severe-risk wards and 93 high- risk wards. The spatial neural estimator achieves a validation MAE of 0.46 risk-score units, while the temporal comparison selects LSTM as the best-performing model with a validation MAE of 1.47 and holdout MAE of 2.00. The results show that interpretable satellite-derived indicators, risk-sector formulas, and validated deep learning models can be combined to support ward-scale climate adaptation planning.

Deep Learning-Based Image Processing of Multi-Sensor Satellite Imagery for Precision Agriculture and Environmental Monitoring: Current Trends and Techniques

2026-09-21T10:37:02+00:00

The recent development of archives of Earth-observation imagery from satellites like Sentinel-1 (Synthetic Aperture Radar) and Sentinel-2 (multispectral optical) has led to the need for automated, scalable image-processing techniques for the extraction of agricultural and environmental information at continental scale. The shift from classical pixel-based and shallow machine learning classifiers to deep-learning networks, like Convolutional Neural Networks (CNNs), encoder-decoder segmentation networks (U-Net and its attention-augmented counterparts), CNN-recurrent networks for multi-temporal sequences, Generative Adversarial Networks (GANs) for super-resolution, and vision transformer networks trained using self-supervised objectives, are reviewed. Based on the latest peer-reviewed literature (2015-2025), the paper synthesizes reported accuracy, F1-score, and Intersection-over-Union trends across different applications of crop classification, land-cover mapping and change detection, and evaluates methodological evolution from single-image classification to the use of multi-sensor pipelines temporally and spectrally fused. Current problems are outlined, such as optical contamination and atmospheric effects; limited and geographically imbalanced labeled datasets; high computational requirements for transformer-based inference; and lack of cross-region generalization. The review suggests that the most viable path forward for operational, weather-resilient agricultural and climate monitoring in the operational phase is using Sentinel-1 SAR with Sentinel-2 optical data through self-supervised and attention-based architectures, especially for regions of the Global South with little data.

Fake News Detection: A Survey of Machine Learning, Deep Learning, and Transformer-Based Approaches

2026-09-24T10:24:48+00:00

Artificial intelligence, social media, and online news platforms have all experienced tremendous growth in recent years, which has led to a considerable increase in the dissemination of fake news. This has resulted in severe problems for public confidence, democracy, healthcare, and economic stability. This is because of the extensive spread of misinformation and disinformation, which makes it increasingly difficult to differentiate between genuine news and content that has been manufactured. As a consequence of this, the detection of fake news through automated systems has emerged as a significant area of research in the fields of artificial intelligence and natural language processing. In this article, a complete overview of various ways for detecting false news is presented. These techniques include rule-based methods, machine learning, deep learning, transformer-based models, large language models (LLMs), multimodal learning, and hybrid approaches. Support Vector Machine (SVM), Random Forest (RF), and Logistic Regression (LR) are examples of traditional machine learning algorithms that offer efficient baseline performance. On the other hand, deep learning models such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM) significantly enhance contextual understanding. The detection accuracy of recent transformer architectures such as BERT, RoBERTa, ALBERT, and GPT has been significantly improved with the implementation of advanced language representation and contextual learning. More robustness and dependability can be achieved by the utilisation of multimodal and ensemble methodologies, which mix information from textual, visual, and social network sources. In spite of these achievements, there are still substantial difficulties that need to be addressed, including multilingual material, misinformation generated by artificial intelligence, limited labelled datasets, explainability, and computational complexity. In light of this, it is recommended that future research concentrate on the development of frameworks for the identification of fake news that are resilient, explainable, and computationally efficient. These frameworks should be able to handle growing misinformation across numerous digital platforms.