We present a day-ahead scheduling strategy for an Energy Storage System (ESS) in a microgrid using two algorithms - Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). The scheduling strategy aims to minimize the cost paid by consumers in a microgrid subject to dynamic pricing. We define an objective function for the optimization problem, present its search space, and study its structural properties. We prove that the search space has a magnification of at least $50\times (B_{c} - B_{d} + 1)$ , where $B_{c}$ and $B_{d}$ are the maximum depths of charge and discharge in an hour (in percentage) of the ESS respectively. In a simulation involving load, energy generation, and grid price forecasts for three microgrids of different sizes, we obtain ESS schedules that provide average cost reductions of 11.31% (using GA) and 14.31% (using PSO) over the ESS schedule obtained using Net Power Based Algorithm.
Aquaculture is a fast-growing food-production sector that accounts for almost 50% of the world’s fish used for consumption. Aquaculture refers to the cultivation of fish in cages. The fish in the cage must be fed at regular intervals, for which the number of fish helps in estimating the amount of feed to be put in the cage. The behavior of fish in a caged environment reflects their health. In the absence of an ambient atmosphere, fish are stressed, which results in frantic movement. The frantic behavior of fish can be identified using recent advancements in image and video processing. In this study, we have focused on frantic behavior detection, fish detection, and counting. For this, a RAS with Tilapia fish has been setup, and the videos of the fish are captured. The detection and counting have been achieved by using the YOLOv5 model. The model has resulted in a precision, Recall and F1-measure of 81%. The results are compared with the ground truth, which indicates that the model has been successful in counting the fish. The frantic movement of the fish has been detected by developing an optical flow model. The results are encouraging and can be used for frantic behavior detection.
Face Recognition is one of the major research areas in Computer Vision. Researchers have applied many image processing techniques and neural networks for the problem but still not able to achieve the desired accuracy for all kinds of data. This work presents a hybrid approach by combining output of two different artificial neural networks PCA-ANN and LDA-ANN. For any given face image, feature extraction techniques have been applied to obtain a representation of the image, using interest point and edge detectors, namely, Harris, SIFT, Canny and Laplacian of Gaussian. Principal Component Analysis and Linear Disciminant Analysis have been actively used for dimensionality reduction of the extracted feature vector. Considering two such different representations, we have trained using an artificial neural network and finally combined the result using a logical OR operation. On Faces94, the proposed approach achieves 98.5% accuracy outshines DeepID and Light CNN-9 approach and fairs significantly better than most state-of-the-art deep learning works.
With the exponential growth of digital data across several domains, multimedia retrieval has emerged as a very critical research area. While much development has been witnessed in feature extraction, indexing methods, and deep learning-based retrieval models, many challenges persistently hamper the creation of efficient and scalable multimedia retrieval systems. This review systematically analyzes recent research in multimedia retrieval, highlighting key methodologies, findings, and limitations across various studies. The major gaps identified include the semantic disparity between low-level features and high-level semantics, as well as challenges of large datasets, privacy and security concerns, and the explainability of deep learning models. Besides, challenges on noisy and imbalanced data, multimodal data sources, and a lack of standardized benchmarking frameworks further limit the performance of existing systems. This paper comprehensively presents these gaps and proposes future research directions to bridge them, thereby directing the development of more robust, scalable, and user-centered multimedia retrieval systems. To overcome persistent challenges in versatility, semantic alignment, and trustworthiness, we introduce a unified framework that seamlessly integrates multimodal semantic fusion, explainable AI, privacy-preserving learning, and continual adaptation for robust and future-ready multimedia retrieval.
This work focuses on comparing the suitability of different machine learning models for the classification of handwritten digits in the Devanagari script. The models that will be compared in this study are: K-Nearest Neighbours (K-NN), Support Vector Machine (SVM), Convolutional Neural Network (CNN), GoogLeNet (Inception v1), and ResNet-50. GoogLeNet and ResNet-50 are complex, deep neural networks. They possess a large number of hidden layers, and are generally used for more complex image classification tasks. The use of these models in this project is to gauge how well they perform on simpler image data. The foundation of this research is based on the ever increasing demand for accurate and efficient digit classification models in India, for purposes such as document scanning, ID card recognition, and the digitization of institutional records. The primary objective of this research project is to identify the most accurate and efficient digit classification model for numbers in the Devanagari script. Surprisingly, proposed simple CNN model outperforms the other complex GoogleNet and ResNet-50 models. Accuracy and Fl score of proposed CNN model is 99.522% and 0.9978 respectively. Also, the proposed CNN model used in this study outperforms other CNN model considered for Devanagari numerals classification.
11 March 2020 in IEEE Transactions on Evolutionary Computation
This paper studies the efficacy of the Metropolis algorithm for the minimum weight codeword problem. The input is a linear code C given by its generator matrix and our task is to compute a non-zero codeword in the code C of least weight. In particular, we study the Metropolis algorithm on two possible search spaces for the problem, the codeword space and the generator space. The former is the space of all codewords of the input code and is the most natural one to use and hence has been used in previous work on this problem. The latter is the space of all generator matrices of the input code and is studied for the first time in this work. In this paper, we show that for an appropriately chosen temperature parameter the Metropolis algorithm mixes rapidly when either of the search spaces mentioned above are used. Experimentally, we demonstrate that the Metropolis algorithm performs favourably when compared to previous attempts. When using the generator space, the Metropolis algorithm is able to outperform the previous algorithms in most of the cases. We have also provided both theoretical and experimental justification to show why the generator space is a worthwhile search space to use for this problem.