‘Good health and well-being’ is the third Sustainable Development Goal formulated by the United Nations. Air pollution is one of the major issues that affects the good health of people across the globe. The quality of air we as ‘humans’ breathe is vital for leading a quality and healthy life. This research attempts to demonstrate the application of deep learning models in the prediction of the air quality index of select Indian cities. This research makes use of the techniques Random Forest, XGBoost, and LightGBM. It has also developed an advanced Deep Learning model for the prediction of the Air Quality Index. The developed models have been tested against a dataset on the air quality of multiple Indian cities. The performance evaluation revealed that the advanced deep learning model with additional layers has rendered better performance than the traditional model Random Forest, XGBoost and LightGBM. The research also gives insights into how deep learning models could be implemented on a real-time basis across all cities in India in order to prevent environmental pollution and ensure sustainable living of human beings on earth.
The ability to recognize human facial expressions is a fundamental requirement for developing advanced human–computer interaction systems. Humans convey emotions primarily through facial expressions, which has driven the development of Facial Expression Recognition (FER) technology in areas such as affective computing and personalized recommendation systems. By leveraging FER, systems can continuously monitor changes in a user’s emotional state throughout the day and generate context-aware responses, thereby enhancing overall user experience. The Spotify application of FER is a music recommendation, where songs are suggested based on facial expressions to match or influence the user’s emotions. Traditional FER approaches primarily rely on spatial-domain features, which often struggle to capture global context and subtle frequency-based patterns in facial images. To address these limitations, this study introduces the Fourier Image Transformer (FIT), a frequency-domain transformer model capable of learning discriminative global representations of facial expressions. By transforming facial images into the frequency domain, FIT captures both local and global variations in facial patterns, improving recognition accuracy and robustness. The proposed framework is trained and tested on the Real-world Affective Faces Multi-Label (RAF-ML) dataset and FER2013, which contains diverse real-world facial expressions under varying conditions. The preprocessing steps, including face detection, image enhancement, resizing, and normalization, to ensure reliable feature extraction. Recognized FER are subsequently mapped to selected music tracks, enabling personalized and image-based music recommendation. This approach offers a seamless, non-intrusive interface that adapts to user emotions in real-time. The FIT model achieved outstanding performance with an accuracy of 98.97%, precision of 98.92%, recall of 98.87%, and an F1-score of 98.90%. The study demonstrates how frequency-domain transformers can enhance FER performance and support intelligent, emotion-driven applications.
This paper develops an image processing-based system in order to verify the contents in each online newspaper. The newspaper contains many sub articles under various headings. Each sub articles in the newspaper may corresponds to various social activities such as Crime, Education, Sports, Politics, Entertainment and Culture. In this work, a newspaper word classification system has been proposed using Deep Learning architecture. This proposed system consists of a preprocessing section, Deep Learning (DL) Architectural module and a data retrieval algorithm. The preprocessing section is used in this work to convert the spatial image into frequency translational image by Gabor transform, and further, Optical Character Recognition (OCR) has been used to transform this frequency translational image into words. The segmented words in the newspaper image are fed into the proposed DL architecture to obtain the classification results. These classification results have been used to find the Euclidean distances with respect to the databases in order to identify the content of the newspaper image.
Human beings can learn throughout their lives; they learn new concepts, refine their existing skills, employ their experiences in various situations, and build memories for prolonged retention. The ability to learn throughout one's lifetime is known as continual learning, and it is challenging to replicate this capacity in an artificial neural network. Deep models tend to overwrite previously learned information when exposed to sequential tasks, a phenomenon known as catastrophic forgetting. To address catastrophic forgetting, several methods have been proposed. In this study, we focus on a comprehensive empirical evaluation of three major categories of continual learning methods: regularization-based methods, replay-based methods, and architecture-based methods. The experiment uses multiple benchmark settings. For a simple image classification task, use Split-MNIST. For representing more complex tasks, use the Split CIFAR10 and Split CIFAR100 datasets (with and without pretraining). All datasets are used under class-incremental and Task-Incremental Learning scenarios. Each scenario faces distinct challenges. To ensure transparency, each method is evaluated using a shared backbone architecture. Performance of each method in both scenarios is reported using various standard continual learning metrics, such as Average Accuracy (ACC), Average Forgetting and Backward Transfer (BWT), along with measurements of total training time and CPU/GPU memory usage to show the resource efficacy. This paper presents a remarkable difference in terms of levels of difficulty and the relative efficacy of various tactics in both scenarios, task-incremental and Class-Incremental Learning. The performance of Task-Incremental Learning is better than Class-Incremental Learning for all continual learning methods.
The slot array antenna is very important in the present communication networks, and it offers high gain and directivity of the radiation pattern. However, in the design, challenges such as low power-handling and directivity are encountered. This paper proposes a novel technique to address these problems by introducing dielectric loading into the slot-array antenna. As the antenna structure, a geometric pattern of slots or openings is cut into a conductive surface. The use of dielectric materials is also carefully considered to enhance power-handling capability and directivity, thereby overcoming impedance-mismatch and radiation-efficiency issues. The optimized design is achieved through an extensive optimization approach, called Slot Array Antenna with Dielectric Loading (SAAbDL), which aims to maximize power handling capacity and directivity. This design uses advanced computational techniques to optimize the arrangement and properties of the dielectric materials in the antenna. Dielectric-loaded prototype slot-array antennas are fabricated using appropriate manufacturing techniques and materials. Efficiency, gain, radiation pattern, and return loss are used to characterize these antennas. The findings of the present study demonstrate the benefits of dielectric loading in increasing the power-handling and directivity of slot-array antennas. The proposed methodology and fabricated prototypes are envisioned to be utilized in high-power communication systems, radar systems, and other application areas where robust, directional radiating characteristics of the antenna are required.
