Deepak Mane

Acute lymphoblastic leukemia (ALL), sometimes referred to as hematopoietic cancer or blood cancer, is a group of cancers that impact lymphocytes, which are white blood cells. Improving patient outcomes and developing efficient treatment plans depend on early and precise blood cancer diagnosis. The lack of labelled data makes it difficult to segment lymphoblast cells from microscopic images. Our research aimed to achieve unsupervised approach for precise and accurate segmentation of blasted lymphocyte cells, thereby improving the overall performance of ALL detection and classification into its subtypes L1, L2 and L3. The proposed method employs k-means segmentation, where the parameter k is tuned, and optimal value is determined based on segmentation quality. For better performance, generated segments are evaluated against ground truth image based on Structural Similarity Index Measure (SSIM), Dice similarity coefficient (DSC) and Intersection over union (IoU). The algorithm iterates over different values of k, assesses the segmentation quality, and selects the segment with the highest evaluation score. Customized convolutional neural networks are employed for categorization. The data augmentation technique has been applied to expand the amount of training data in order to enhance model efficiency. The ALL-IDB dataset is used to assess the model's performance, and the experimental results showed that the suggested method can identify blasted cell subtypes with an overall accuracy of 99%. We succeeded in detecting acute lymphoblastic leukemia with 100% accuracy. Our proposed model not only enhances the accuracy significantly but also determines the optimal value of clusters (k) for more effective segmentation.

Natural language, sign language employs several modes of expression for everyday communication. Researchers have paid less attention to ISL interpretation than other forms of sign language. This paper introduces an automatic technique for translating manual letter movements into Hindi sign language. Because it works with pictures of bare hands, the user may interact with the device naturally. System gives deaf individuals the chance to interact with hearing people without the use of an interpreter. This paper goal is to develop techniques and a system for automatically recognizing Hindi sign language. The creation of a Hindi Sign Language database is the initial stage in this system. The most important stage in any hand gesture recognition system is hand segmentation since higher recognition rates may be attained with better segmented output. To improve identification rates, the suggested system additionally incorporates reliable and effective hand segmentation and tracking algorithms. Using a large sample set, it has been shown to be feasible to recognise 43 different words from the Standard Hindi sign language. We want to be able to convert certain very basic parts of sign language into text in the suggested system and vice versa in Hindi. In this paper, we present a number of new features and prospective advancements:The emphasis on real-time systems that recognize and interpret sign language motions from live video input is one noteworthy advance. With the use of this feature, people who use sign language and others can communicate instantly without any delays or middlemen. Integration of Speech Synthesis and Hand Gesture Recognition: This innovative approach to communication accessibility combines speech synthesis and hand gesture recognition technology. The paper provides a comprehensive solution for promoting engagement and understanding between the deaf community and others who communicate primarily through spoken language by translating gestures from sign language into intonation-rich speech.

Sentiment analysis is an essential part of gaining deeper insights into particular tweets. Here, we analyse the inclinations that reflect user opinions and their sentiments regarding the tweets about the Russia and Ukraine war. Taking data from social media, we use sentiment analysis to gain insight into the wide range of public opinion. In this paper, we proposed customized Bidirectional Encoder Representations from transformers (BERT) natural language processing model to gauge the tone of the tweets about the war. A proposed model for sentiment analysis is fine- tuned on a dataset related to the Russia- Ukraine war. Proposed model to classify sentiments in domain specific texts to ensure accurate and context aware sentiment predictions. Our unique approach leverages BERTs contextual language understanding model to study Russia- Ukraine relations. The model achieved an accuracy of 94.73% on the fitted dataset. Everyone benefits from this research, and the public as it gives them a better understanding of public opinion. Experiment results shows that the performance of the proposed model is better than the existing approaches.

In today’s modern era Email spoofing is a serious danger to email communication because it allows hackers to impersonate legitimate senders and change email headers like the email ID and IP address. And email headers are vital parts of any emails that carry important details about the sender, recipients, email delivery method, and email content. These headers are essential for email management and communication since they fulfill a variety of functions, such as email filtering, security, and traceability. So, if the non-legitimate user sends an email then mainly any of the important fields in headers is mismatched. So the proposed mechanism in this paper provides a reliable email spoofing detection at the server level, utilizing cutting-edge cyber security practices and artificial intelligence, in response to this expanding concern.

Today, the importance of regional language translation has grown significantly as it facilitates effective communication, accessibility, and inclusivity for diverse populations in a globalized world. It enables individuals to access information and services in their native languages, fosters cultural preservation, and enhances opportunities for education and economic growth. So, this paper presents a thorough implementation of the Transformer-based neural network architecture for regional language translation. Our methodology enables translation between languages by utilizing cross-attention and the self-attention mechanism. Because the architecture is modular in nature, it is simple to configure model parameters such as dropout rates, model dimensions, and layer count. We present feed-forward layers, layer normalization tokenization, positional encoding, and multi-head attention components. The Transformer model, when used in both the encoder and the decoder, may acquire intricate linguistic connections and patterns. We showcase the model's potential for natural language processing tasks by demonstrating its effectiveness in translating text between language with the help of Samanantar Dataset, a parallel corpus containing a total of 4,000,000 English and Kannada sentence pairs. The results showed a consistent rise in training accuracy with each epoch, highlighting the model's capacity to learn and generalize from the data, with notably enhanced performance correlated with larger datasets and more potent hardware utilization. This work offers a flexible foundation for future research in the subject and advances the creation of cutting-edge machine translation models.

This paper introduces two novel class-specific fuzzy clustering algorithms: Mean-based Supervised Clustering (MSC) and Density-based Mean Supervised Clustering (DMSC). These algorithms are designed to construct the hidden layer of the Fuzzy Hypersphere Neural Network (FHNN) classifier, which is structured on the framework of the Radial Basis Function Neural Network (RBFNN). The FHNN classifier utilizes fuzzy sets as labeled pattern clusters in its hidden layer, with classes represented in the output layer formed by the aggregation of these fuzzy sets. An important characteristic of this classifier is its independence from tuning parameters. It meticulously determines centroids and radii for labeled clusters, consistently achieving 100% accuracy across any training set. The FHNN classifier effectively handles outliers and is robust to variations in data presentation, ensuring clear data visualization for users. During the creation of labeled clusters in the hidden layer, binary weight values are adjusted concurrently between the hidden and output layers. This study proposes the formation of fuzzy clusters with varying dimensions tailored to the dataset. The classifier architecture, rooted in the radial basis function neural network, achieves 100% training accuracy due to precise fuzzy cluster formation. Experimental comparisons with RBFNN and similar classifiers using sixteen benchmark datasets demonstrate the superiority of the proposed classifier in pattern recognition tasks.

There have been significant advancements in object detection in recent years, notably with the emergence of the YOLO approach. In this paper, we proposed a thorough comparison between YOLO-NAS and YOLOv8 specifically for detecting vehicle license plates based on the analysis of experimental results. Real-time vehicle license plate detection can have a wide range of applications like crime prevention, surveillance, traffic-pattern analysis, automated toll collection, etc. Our evaluation uses a diverse dataset capturing vehicle images from various angles and levels of occlusion to ensure a comprehensive assessment of the models' capabilities. The metrics considered include precision, recall, and mAP, offering insights into detection accuracy and false positive rates. The results reveal nuanced performance differences between YOLO-NAS and YOLOv8 in vehicle license plate detection tasks. YOLO-NAS demonstrates superior accuracy in specific scenarios due to its optimized architecture. Conversely, YOLOv8 shows notable efficiency gains, especially regarding inference speed, making it a compelling choice for real-time applications. Here, we contribute valuable insights to the ongoing discussion on object detection methodologies, assisting practitioners in making informed decisions when selecting a model for vehicle license plate detection applications. The findings underscore the importance of considering trade-offs between accuracy and computational efficiency in specific use cases

Diabetic Retinopathy threatens vision in diabetics, necessitating swift and accurate detection. This study employs Convolutional Neural Network (CNN), ResNet50, and InceptionV3 for automatic DR identification, achieving a notable 96.18% accuracy over 80 epochs. To enhance robustness, a pre-processing pipeline incorporates Gaussian filtering, CLAHE, median filtering, and top-hat filtering, significantly advancing DR detection accuracy. Evaluation on the APTOS 2019 dataset (1299 training, 279 testing images) reveals great accuracy as well as sensitivity, and specificity, forming a basis for early intervention and vision impairment prevention. This research at the nexus of DL which is also known as deep learning and medical image analyze offers a promising solution for early DR detection. The 96.18% accuracy demonstrates practical viability, positioning our approach as a valuable tool for healthcare practitioners and ophthalmologists in effectively diagnosing and managing diabetic retinopathy.