Deep learning techniques have transformed image processing by providing powerful tools for extracting detailed patterns and features from large amounts of data. This paper provides a unique hybrid methodology for hyperspectral image (HSI) analysis that integrates Random Forest (RF) feature selection with Convolutional Neural Networks (CNNs) for classification in the domain of HSI classification. The study leverages CNNs' inherent automatic hierarchical feature extraction ability and RF's effectiveness in recognizing and conserving essential features. This study thoroughly validated the proposed approach using the Indian Pines dataset, a prominent HSI dataset of dimensions 145x145 pixels and 200 spectral bands collected by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). The combination of RF-driven dimensionality reduction and CNN-based classification yields a robust model with an accuracy rate of 99.35%, demonstrating its efficacy in categorizing HSIs with varying object scales.

Communication for individuals who are hearing- and speech-impaired, commonly referred to as the deaf and mute community, heavily relies on sign language as their primary mode of expression. This study presents a novel framework leveraging image processing techniques for the detection and recognition of sign language gestures. The developed software offers promising avenues for enhancing comprehension of Sign Language, with potential applications in educational settings, public spaces, and interpersonal interactions. The proposed method streamlines the recognition process of sign language, employing deep learning algorithms for the accurate prediction of signs. The system operates by processing input images containing signs through a convolutional neural network, encompassing stages such as pre-processing, feature extraction, model training, testing, and sign-to-text conversion. Crucially, the system's output provides text-based descriptions of the sign in the input image and notably integrates voice output for enhanced accessibility and communication. This multifaceted approach contributes towards bridging communication barriers between individuals with disabilities and those without, promoting inclusivity and understanding in diverse social contexts.

The development of a Smart Guidance Stick, an innovative technology solution aimed to increase mobility and independence of people with visual impairments, is discussed in this project. The Stick incorporates sensors, microcontrollers, communication tools, and new features such as real-time smart support and emergency contacts. Its goal is to offer visually impaired persons the confidence to navigate unfamiliar surroundings, detect obstacles and get precise instructions via audio guidance. It also has an emergency alert system in case of an emergency. This innovation addresses barriers faced by the blind and visually challenged community by combining sensor technology, algorithms, and communication tools to give greater navigation and autonomy. This work demonstrates how hardware features and components can be successfully integrated.

VANETs are wireless communication networks that enable vehicles to communicate with each other and RSUs. RSUs are crucial in establishing connectivity among the moving vehicles on the road and enhancing road safety and traffic management. For operational purposes, RSUs require electrical energy which is expensive, hence it is highly advisable to prioritize the preservation of this valuable energy resource. Therefore, the energy consumption across all RSUs should be minimized for efficient usage. Apart from operating itself, RSUs need energy to fulfill the requests they receive. These requests arrive at different time stamps and require the RSUs for different intervals of time for their processing. Hence, scheduling of these requests should be done so that the energy of the RSUs does not get under or over utilized. To do so, we have applied Median Average Round Robin algorithm (MARR) and Round Robin algorithm (RR) on the VANET scenario. For comparative analysis, the results have been compared with a random scenario with no scheduling algorithm and also with the nearest fast scheduling algorithm (NFS). It turns out that MARR_VDS performs the best by reducing the energy usage by 50% approximately and also minimizing the overall cost of VANET by using 55% lesser number of RSUs.

The Internet of Things (IoT) networks always operate within the context of diverse and constrained characteristics of the devices. Low-Power and Lossy Networks (LLNs) constitute a network architecture commonly utilized in IoT application deployments, facilitating networking and the establishment of paths for data transmission. The Routing Protocol for Low-Power and Lossy Networks (RPL) demonstrates promising capabilities for LLN network operations, supporting IPv4 and IPv6-enabled services. The RPL protocol constructs a Destination Oriented Directed Acyclic Graph (DODAG) logical routing topology based on defined Objective Function (OF) metrics. Routing operations within the DODAG utilize these metrics and constraints to select parent nodes and calculate optimal routes between two nodes. Standardized OFs have traditionally focused on either parent node selection or routing objectives within the DODAG, often treating load balancing and bottleneck optimization separately. However, their combined impact on RPL's effectiveness has been overlooked. This paper introduces an Adaptively Composite Objective Function (AC-OF) approach that considers the combined objectives of DODAG load balancing and optimized routing operations. Through simulation evidence, the paper presents improved network parameters. The AC-OF implementation brings out significant results in the form of a balanced DODAG topology and it has good impacts on data transmission, control overhead messages, parent switching, delay, energy consumption, and node lifetime.

One of the most popular OSs utilized by the public these days is Windows. A serious concern to the security and integrity of Windows OS systems is the proliferation of malware. The goal of this research project is to create a practical method for identifying and categorizing various malware kinds on the Windows operating system to combat the pervasive malware problem. For the efficient identification and classification of malware on Windows, an ensemble technique using hybridization of Support Vector Machine, Decision tree, and Logistic Regression is proposed. The suggested method makes use of the idea of feature selection methods to determine the patterns and signatures of numerous malware families. The genuine malware dataset will be used to test and assess the suggested ensemble as well as the current basic machine learning techniques. In the end, this will assist both the novice and the expert in cyber security to comprehend and prepare for the ever-changing threats posed by the new breed of malware on Windows PCs.

A novel method is proposed to augment the precision and efficiency of the fatigue reliability analysis. This method utilizes an Improved Northern Goshawk Optimization (INGO) algorithm to optimize the BP neural network (BPNN). The enhanced Circle Chaotic Map, adaptive inertia weight, Elite opposition-based learning strategy, and artificial rabbits optimization strategy are incorporated to enhance both the accuracy and speed of the optimization algorithm. Proposing a fatigue reliability calculation approach for the bogie frame based on the maximum material utilization rate. The approach computes fatigue reliability through the segmentation of all bogie frame welds, employing the maximum material utilization rate within each experimental design scheme. The INGO-BPNN is trained using 80 sample sets acquired through experimental design. The fatigue reliability of the bogie frame is assessed using the Monte Carlo method based on the surrogate model. Research findings indicate that the surrogate model achieves a predictive accuracy of 99.8%, while the fatigue reliability of the bogie frame stands at 99.36%. The proposed method improves the calculation efficiency while ensuring the prediction accuracy of the model.