Precise prediction and detection of epilepsy, a common neurological condition marked by sporadic and repeated seizures, is necessary for efficient management and patient care. To do accurate and reliable seizure detection, machine learning (ML) on electroencephalogram (EEG) recordings is a good choice because of its capacity to provide insightful information about brain activity during seizures. In this work, we do a thorough comparative analysis of five machine learning methods for the prediction of epileptic seizures using EEG data: Random Forest (RF), Decision Tree (DT), Extra Trees (ET), Logistic Regression (LR), and Gradient Boosting (GB). Firstly, to assure the quality of the data and enable precise model training, the dataset underwent extensive preprocessing, which included cleaning, normalization, treatment of outliers, and oversampling. These preprocessing techniques significantly enhanced the models' performance. In our study, we evaluate the accuracy of each model; the LR classifier obtains 56.95% accuracy, while both GB and DT achieve 97. 17% accuracy. Outperforming the findings of previous investigations, the ET model showed the best performance with an accuracy of 99.29%. Achieving 98.99% accuracy was another accomplishment for the RF model. Our results imply that the ET model not only performed better than the other models in the comparison analysis but also more effectively than the most advanced outcomes from earlier research. Regarding the accurate and dependable prediction of epileptic seizures using EEG data, the ET model is a solid option due to its remarkable performance.
The myoelectric prosthetic devices advancement has been essential to redevelop the grasping capabilities of amputees. Despite these advancements, myoelectric prosthetic devices need improvements to replicate the grasping performed by the human hand. The grasping performed by the human hand needs to be firm and avoid slippage of the objects. To avoid slippage, the information related to grasping force is important. In this study, the EMG signals are acquired while grasping a cylindrical object at different weight levels with two precision prismatic gestures. Using these EMG signals, force-based classification is performed based on the different weight levels for each gesture. The result shows that the highest mean classification accuracy was obtained using Support Vector Machines (SVM), followed by the k-Nearest Neighbors (k-NN) for each gesture. The mean classification accuracy obtained using SVM were 94.05% and 96.8% for 1st and 2nd gesture respectively. It is concluded that better outcomes are obtained using more complex classifiers as compared to simple classifiers such as Naïve Bayes and Linear Discriminant Analysis. In the future, a more descriptive and detailed analysis is expected to be performed using the outcomes obtained.
The tremendous volume of video data generated by industrial surveillance networks presents a number of difficulties when examining such videos for a variety of purposes, including video summarization (VS), analysis, indexing and retrieval. The task of creating video summaries is extremely difficult because of the huge amount of data, redundancy, interleaved views and light variations. Multiple object detection and identification in video is difficult for machines to recognize and classify. To address all such issues, multiple low-feature and clustering-based machine learning strategies that fail to completely exploit VS are recommended. In this work, we achieved VS by embedding deep neural network-based soft computing methods. Firstly, the objects in extracted frames are detected using YOLOv5, and then the frames without objects (useless frames) are removed. Video summary generation occurs with the help of frames containing Objects. To check the quality of the proposed work Summary length, precision, recall, PR curve, and mean average precision (mAP) are used and system resource utilization during the model training are also tracked. As a result, the proposed work was able to identify the most effective video summarization framework with best summary length under varying conditions.
In recent years, the significant research area in the domain of multimedia security is image information security. This can be provided by converting it into a secret code through the process of encryption or hiding its true significance or a combination of both. It is required in many domains like defence, military, banking, education, etc, to keep these sensitive images from being hacked or destroyed. In this paper, BEMSS, an (n,n) multi-secret image encryption with a secret sharing scheme is proposed using blockwise encryption. In the proposed technique two-level encryption, modulo and inter blockwise encryption, are performed to the n secret images, and a two-level share generation process is performed to the encrypted images. BEMSS recovers n lossless the secret images by modulo and blockwise decryption process. Experimental results obtained show the effectiveness of the proposed BEMSS which provide a computationally efficient security scheme.
In growing electricity markets, there is a demand for more effective, secure, and pure sine wave inverters due to the fast expansion in the use of solar photovoltaics in renewable power generation. Despite the fact that certain inverters are affordable price and have better efficiency, they fail to reduce the Total Harmonic Distortion (THD) of their output voltage and load current, and also lack galvanic isolation between source and load. This paper aims to design and simulate the 9-Level Cascaded H-Bridge Multilevel inverter with Low THD and Galvanic Isolation (CHB MLI) and to compare the results with the conventional two-level inverter output. Also, to develop the proposed MLI topology in hardware using Field Programmable Gate Array (FPGA) technology and to develop a Matlab application for real-time controlling and monitoring.
