Natural Language Processing (NLP) primarily focuses on understanding and processing human language in a manner that machines can comprehend. NLP covers a wide range of tasks that enable machines to interpret and generate human language. These include identifying entities in text (Named Entity Recognition), answering questions, summarizing documents, classifying text, extracting and retrieving information, translating between languages, analyzing sentiment, and recognizing spoken language. Machine Translation (MT) has evolved from rule-based systems in the 1940s to sophisticated neural architectures that achieve near-human performance on high-resource language pairs. However, significant challenges remain in low-resource languages, domain adaptation, and evaluation standardization. This systematic review synthesizes 77 years of MT research to (1) quantitatively compare the evolution of MT approaches, (2) identify performance trends across different methodological paradigms, (3) analyze evaluation metrics and benchmark datasets, and (4) characterize research gaps, particularly for low-resource languages, including Indian regional languages. Following the PRISMA guidelines, we systematically searched eight academic databases (IEEE Xplore, ACM Digital Library, ACL Anthology, SpringerLink, arXiv, and others) from 1948 to 2025. Our search strategy included controlled vocabulary and Boolean operators, resulting in 1,354 initial results. After applying the inclusion/exclusion criteria and quality assessment, 164 studies were selected for final analysis. We extracted quantitative performance metrics and methodological details and conducted a comparative synthesis of Statistical Machine Translation (SMT), Knowledge-Based Machine Translation (KBMT), Neural Machine Translation (NMT), and hybrid approaches. Several researchers in the machine translation domain have attempted to translate sentences of different languages, views of newspapers, selections of epics, children’s stories, and novels into English, and translations of simple English sentences into Indian languages. However, they have not found sufficient accuracy in the translation of uncommon words, phrases, idioms, and sayings, or proverbs of Indian and non-descript proverbs. Thus, there is a major necessity to develop a viable machine translation scheme to translate several Indian and foreign languages (including Arabic, German, Japanese, French, and Russian) into English and vice versa. The field requires standardized evaluation protocols, expanded multilingual benchmarks (particularly for Indian languages, where only 23 studies exist), improved reproducibility practices, and a focus on practical deployment challenges. Our comprehensive benchmark comparison and identified research gaps provide a roadmap for future MT development, emphasizing the critical need for low-resource language support and robust evaluation frameworks.

Bird image segmentation and classification from natural-scene images captured by a device are important for ecological research focused on wildlife conservation and monitoring. Manually segmenting a bird image from a complex background is tedious, time-consuming, and process-intensive. Therefore, an automatic segmentation of bird images from the complex scene image, exploiting image processing techniques, has been extensively studied. Although many efforts have been made to propose segmentation techniques in a different context, the techniques cannot be generalized and may not perform well in some cases. Besides, the task of segmentation is very challenging due to occlusion, morphology, and lighting conditions. Therefore, there is a scope to fine-tune and enhance the segmentation ability of algorithms. The present study proposes a novel method of ensembling deep learning models using Genetic Algorithm (GA) for improving bird image segmentation in a complex background. The proposed methodology employs the segmentation results obtained by the individual deep learning models at the pixel level for ensembling. Five popular deep learning models, namely Unet, PSP-net, Link-net, Feature Pyramid Network (FPN), and Deeplabv3+, are used in the study. The CUB-200-2011 benchmark dataset was employed to conduct experiments, and the efficacy of the proposed approach was evaluated using the Intersection over Union (IoU) metric. The segmentation results obtained were very promising, achieving the highest IoU compared to contemporary segmentation approaches.

The precise identification of thoracic disorders using chest X-rays (CXRs) is essential for favorable clinical diagnosis. However, it continues to be difficult because of inter-class similarities and overlapping symptoms. This study introduces a hybrid MobileNetV2–Vision Transformer (ViT) framework for multi-label classification of CXR images into 14 disease categories on NIH CXR14 dataset. MobileNetV2 serves as a lightweight feature extractor to obtain distinctive spatial representations that are tokenized and enhanced with positional encodings. The encoded features were processed via transformer layers, wherein self-attention mechanisms captured long-range interdependence and global contextual relationships. An output layer based on the sigmoid function facilitated the concurrent prediction of various illness labels. The novelty of the proposed work lies in the PSO-tuned hybrid architecture, which adaptively optimizes key hyperparameters of the MobileNetV2–ViT framework to achieve improved accuracy, faster convergence, and enhanced computational efficiency compared to conventional approaches. This implementation uses an economical, GPU-accelerated instance tailored for inference workloads. It is equipped with a single NVIDIA A10G GPU with 24 GB of dedicated GPU memory, and the proposed work is carried out on AWS SageMaker. The experimental assessment employing five-fold cross-validation showed strong performance, with an average accuracy of 94.6%, precision of 88.1%, recall of 85.9%, specificity of 94.17%, and F1-score of 91.1%. Grad-CAM representations increased interpretability by identifying disease-relevant areas. The results underscore the efficacy of the proposed hybrid technique as a dependable instrument for computer-aided diagnosis of thoracic illnesses.

In digital imaging for medical diagnostics, especially chest X-rays, raster images like JPEG, PNG, and TIFF are frequently utilized. For effective preprocessing, annotation, and machine learning training, large-scale image collections must be arranged according to format and resolution. This study suggests a scalable method for sorting and storing raster-type medical images using Elastic MapReduce (EMR) from Amazon Web Services (AWS). The pipeline uses AWS S3 storage and the Hadoop MapReduce architecture to distribute the identification of image characteristics and arrange them into structured S3 pathways. The outcomes show fault tolerance, cost-effectiveness, and high throughput for datasets with more than hundreds of thousands of images. Elastic MapReduce has gained popularity as a framework for handling massive amounts of data because of its fault-tolerant, scalable, and economical infrastructure. This study examines optimization strategies, assesses performance under various workloads, and looks into the integration of image processing pipelines into EMR clusters. The findings demonstrate that EMR can significantly increase throughput and scalability for large-scale image processing activities as classification, feature extraction, and filtering.

In this work, clustering based one shot approach for segmentation of bird images from complex background using Deeplabv3+ is proposed. Hierarchical clustering technique employed with average linkage is used to obtain clusters of bird images consisting of a particular class. The concept of clustering is exploited to capture intra-class variations and to train the learning model more efficiently and more effectively. Mobilenetv2 architecture is explored to extract features from each bird image, which are used for clustering. After obtaining clusters for each bird's species (class), one representative bird image from each cluster is considered for training the model. The remaining samples in the cluster are used for testing the model's performance in segmenting the bird image from complex background accurately. The proposed segmentation approach is tested on the benchmark bird image dataset called CUB-200-2011 dataset. The experimental results have proved that the proposed cluster-based one shot Deeplabv3+ approach has produced good results with less training.

Advancements in Natural Language Processing (NLP) driven by machine learning, deep learning, and artificial intelligence have significantly broadened its scope and improved interactions between humans and computers. Despite these advancements, NLP systems encounter challenges arising from incomplete and error-prone data, which can result in biased model outputs. Technical domains present further hurdles, necessitating domain-specific fine-tuning and the development of custom lexicons. Additionally, many languages lack robust NLP support, limiting accessibility. In this context, innovative NLP data pre-processing and tokenization methods tailored for Kannada Vachana Sahitya texts are investigated. The majority of documents or text utilized in any language processing applications consists of raw text, in which some of the words are not represented in the standard form. There is a necessity of pre-processing the input text before building the translation model. In this paper, the main goal is to pre-process the source text with essential steps like Cleaning, Parsing, Tokenization, Padding, Stemming and Lemmatization. During this process, Kannada source text - Vachana Sahitya is tested on bilingual corpus and approximately 35% of non-standard input text is considered for experimental analysis.

This paper presents a hybrid approach for the design and development of Lemmatizer for Kannada Vachana Sahithya. Kannada has a rich literary history and is officially recognized as one of the India's classical languages. The existing work on Kannada Morphological Analysis and Lemmatization is conducted on Kannada nouns and verbs from “Kannada Shabdha Kosha” and Kannada text from Government Documents, Circulars and news from All India Radio. In this paper, a hybrid lemmatizer for Kannada Vachana Sahithya is designed and evaluated on an annotated dataset of 5,735 words. In 5735 words of KVS, 80% (4588) are considered for training and 20% (1147) words are used as testing data. The experimental results with the hybrid lemmatizer demonstrate that the proposed model achieves approximately 90% accuracy across various types of KVS input text.

This paper presents a hybrid approach for the design and development of Lemmatizer for Kannada Vachana Sahithya. Kannada has a rich literary history and is officially recognized as one of the India's classical languages. The existing work on Kannada Morphological Analysis and Lemmatization is conducted on Kannada nouns and verbs from "Kannada Shabdha Kosha" and Kannada text from Government Documents, Circulars and news from All India Radio. In this paper, a hybrid lemmatizer for Kannada Vachana Sahithya is designed and evaluated on an annotated dataset of 5,735 words. In 5735 words of KVS, 80% (4588) are considered for training and 20% (1147) words are used as testing data. The experimental results with the hybrid lemmatizer demonstrate that the proposed model achieves approximately 90% accuracy across various types of KVS input text.

Rice blast disease is strongly dependent on environmental and climate factors. Integration of rice blast disease severity prediction model based on climate factors provides decision support framework for farmers to overcome from the problems of climate change scenarios. Major contribution of the proposed study is to predict the severity of rice blast disease using Linear SVM model. Prediction of severity of rice blast disease is Severity of rice blast disease is divided into four classes 0,1,2 and 3. Data imbalance is the most difficult problem in multi-class classification. Proposed study has handled imbalanced data efficiently using k-means SMOTE and SMOTE over sampling techniques to make training and testing data balance. Finally cross location and cross year models are developed using linear support vector machine and predicted severity of rice blast disease to the classes 0,1,2,3 respectively. Cross year and cross location models are cross validated using 5-fold cross validation.

Soil nutrient is an important aspect that contributes to the soil fertility and environmental effects. Traditional evaluation approaches of soil nutrient are quite hard to operate and they are very slow, making great difficulties in practical applications. The proposed study, presents extreme learning machine (ELM) for analysing the soil fertility index values of boron, zinc, organic carbon and pH in Davangere District. Boron, zinc, organic carbon, and pH concentrations in soil play significant roles in paddy crop cultivation and growth. Proposed ELM-based approach helps in the prediction of boron, zinc, organic carbon and pH index values in soil by evaluating four linear and nonlinear activations functions. Performance of ELM model is analysed by increasing the number of hidden neurons in the hidden layer.

This work extricates the image characteristic features for the classification of archeological monument images. At the pre-processing stage, archeological dataset sample images are treated by using structure safeguarding image abstraction framework, which can deliver the most effective image abstraction output by manipulating the perceptible features in the given low-illuminated and underexposed color image samples. Proposed abstraction-framework effectively boosted the significant image property features like color, edge, sharpness, contrast and suppresses complexity and noise. The image properties were also refined at each phase based on the attained statistical feature disposal information. The work adopted the Harris feature identification technique to identify the most significant image features in the input and enhanced images. The framework also preserves significant features in the foreground of an image by intelligently integrating the series of filters during rigorous experimental work and also diminishes the background content of an input image. The proposed archeological system evaluates every stage of the result with assorted subjective matters and calculates the image quality and properties assessment statistical attributes. By this way prominent features in an image have been recognized. The efficiency of this work has been corroborated by performing the trials on the selected archeological dataset. In addition, user’s visual feedback and the standard image quality assessment techniques were also used to evaluate the proposed pre-processing framework. Based on the obtained abstraction images from the framework, this work extracts the image gray color texture features using GLCM, color texture from CTMs and deep-learning features from AlexNet for the classification of archeological monument classification. This work adopted a support vector machine as a classifier. To corroborate the efficiency of the proposed method, an experiment was conducted on our own data set of Chalukya, Kadamba, Hoysala and new engraving monuments, each domain consisting of 500 archeological data set samples with large intra-class variation, with different environmental lighting condition, low-illumination and different pose. Implementation of this work was carried out in MATLAB-2020 with HPC Nvidia Tesla P100 GPU, and obtained results show that combination of multiple features significantly improves the performance to the extent of 98.10%.

Rice blast disease is strongly dependent on environmental and climate factors. This paper demonstrates the integration of a rice blast disease severity prediction model based on climate factors, providing a decision-support framework for farmers to overcome these problems. The major contribution of the proposed study is to predict the severity of rice blast disease using the linear SVM model. Prediction of rice blast disease severity is divided into four classes: 0, 1, 2, and 3. Data imbalance is the most challenging problem in multi-class classification. This study has efficiently handled imbalanced data using k-means SMOTE and SMOTE oversampling techniques to balance training and testing data. Finally, cross-location and cross-year models are developed using a linear support vector machine and predict the severity of rice blast disease to the classes 0, 1, 2, 3, respectively. Cross-year and cross-location models are cross-validated using five-fold cross-validation.

In recent years, the application of artificial intelligence (AI) in agriculture has grown to be the most important research domain. The proposed work focuses on forecasting rice blast disease outbreaks in paddy crops. Disease management in the farm fields is the most difficult problem on the planet. There is a variety of reasons for this, first, a lack of farmers’ experience in diagnosing diseases, second experts’ experience in detecting diseases visually, and third unfavorable climate. In recent days, researchers have offered a variety of time-series techniques in different applications. This study adds time-series techniques to the field of agriculture by forecasting crucial rice blast disease outbreaks in the paddy crop of the Davangere region based on daily weather data obtained from KSNDMC. The statistical time-series technique called ARIMA is trained by employing real data of blast disease outbreaks in the Davangere region from the period of 2015–2019. Meanwhile, the deep BiLSTM model is trained by employing real weather data and blast disease outbreaks of the Davangere region. Both models are evaluated by performance metrics, such as mean squared error and mean absolute error. The proposed research is focused on the hybrid model ARIMA–BiLSTM which is a combination of the statistical ARIMA model and deep BiLSTM model. The seasonal component of the rice blast disease outbreak feature is extracted from the additive decompose function used in the ARIMA model and fed as a dependent feature for the BiLSTM model. According to the results obtained, the hybrid approach can successfully forecast blast disease outbreaks in paddy crops with a mean squared error of 0.037 and a mean absolute error of 0.028 compared to the statistical ARIMA and deep BiLSTM model.

Due to the evolution of computer vision and non-photorealistic rendering (NPR) techniques, enhancements of image features from blatant range images are possible. Conveying the shape from blatant range images are a crucial part of surveillance systems, content-aware analysis, image abstraction and line drawing. In view of this, the work presents a combinational high dynamic range (HDR) and image abstraction framework that can deliver the most effective dirt-free line drawing output to convey the shapes from blatant range images. The proposed framework manipulates the visual features from under/overexposed 2D blatant range images by retaining the prominent tonal information, dominant structural features and suppressing the superfluous details. Significant image properties and quality assessment metrics are effectively enhanced based on the statistical parameters computed and by empirically defined conditions at every stage of the framework. The framework exploits image and objective spatial data to create the dirt-free line drawing in order to recognize amplified elements of the enhanced structure by making use of the Harris key-feature detector algorithm. A sequence of HDR tone mapping operators and NPR image filters are comprehensively integrated through rigorous experimental analysis. Hence, this work empirically retains the prominent tonal and structural features in the frontal region and diminishes the background features in given input images. The work is implemented in MatLab-2020 with a 6.6 teraflops/s high-performance super computation ambience and Tesla P100 graphical processing unit. Efficacy of the presented framework has been validated by executing extensive experimentation on the benchmark datasets such as Ruixing Wang dataset, Flickr repository images and many other interesting datasets are collected. The obtained results are compared with other comparable existing work cited in the literary-works. Furthermore, human visualization perceptual analysis opinion process is also used to evaluate the proposed framework. Significant image abstraction and dirt free line drawing detentions, design challenges, applications and potential work in the domain of non-photorealistic rendering are also envisaged in this paper.

Among various diseases of paddy affecting rice production and cultivation, blast majorly called as rice blast disease has the predominant impact. Thus, monitoring and early prediction of the occurrence of rice blast disease are very important and would be largely helpful for prevention of blast disease. Here, we have proposed LSTM and MLP based machine learning models for rice blast disease prediction and prevention. Historical seven metrological data are used to make prediction of blast disease, two days before its actual occurrence. According to the literature survey conducted in this study, we have made an observation that rice blast disease would outbreak when Minimum Temperature is between 20-26C and Maximum Relative Humidity is ≥90%, hence region specific models are developed for four regions of Davangere district: Chanagiri, Davangere, Harihara, Honnalli. We have adopted curve shift method and two more user defined functions namely temporalize and scale in LSTM model. Performance of the proposed models are evaluated considering classification metrics such as accuracy, precision and recall. In the study conducted, dropout rate is varied from 0.1 to 0.9 for LSTM model and number of hidden layer are added from 1 to 4 in MLP model. For all the regions, both LSTM and MLP model predictions are accurate, and compared to LSTM model, performance of MLP model accuracy is high. These models will be very helpful for rice cultivator and researchers than using regular blast disease prediction model

The proposed abstraction framework manipulates the visual-features from low-illuminated and underexposed images while retaining the prominent structural, medium scale details, tonal information, and suppresses the superfluous details like noise, complexity, and irregular gradient. The significant image features are refined at every stage of the work by comprehensively integrating a series of AnshuTMO and NPR filters through rigorous experiments. The work effectively preserves the structural features in the foreground of an image and diminishes the background content of an image. Effectiveness of the work has been validated by conducting experiments on the standard datasets such as Mould, Wang, and many other interesting datasets and the obtained results are compared with similar contemporary work cited in the literature. In addition, user visual feedback and the quality assessment techniques were used to evaluate the work. Image abstraction and stylization applications, constraints, challenges, and future work in the fields of NPR domain are also envisaged in this paper.

This work identifies the strong dominant features by its location and extracts the image features for the purpose of automatic desire focusing on prominent structure and artistic stylization of images. At the pre-processing level, dataset image is treated using refined structure preserving image abstraction framework which can deliver the best effectual structure preserved abstracted results by utilizing visual attributes from 2D color image. The presented framework efficiently conserves the structural characteristics in the foreground of an input image by exhaustively amalgamate the series of non-photorealistic rendering image filters over meticulous investigational work and it also reduces the background substance of an image. The framework assesses image and object space details to generate structure preserved image abstraction thus distinguishing the accentuated elements of an enhanced structures using Harris key-point feature detector and chooses the 100 major unique dominant feature locations among available features. This work automatically selects the unique location from the extracted features using polynomial region of interest and unselected image regions and its background are blurred using Gaussian motion blurring with point spread function. Deblurring the selected region using wiener filtering to get the desire focusing on prominent structure followed by color quantization and flow-based bilateral filtering is applied over focused structural region to achieve artistic stylization. Efficiency of the framework has been validated by carrying out the trials on the selected Flickr repository, David Mould and Ruixing Wang dataset. In addition, user’s visual opinion and the image quality estimation methods were also utilized to appraise the proposed pre-processing framework. This work lists the structure preserving image abstraction framework applications, limitation, execution difficulties and future work in the field of Non-photorealistic rendering domain.

This work presents a structure-preserving non-photorealistic rendering (NPR) framework that can produce an effective structure-preserving abstracted and stylized output by manipulating visual features from 2D color image. The proposed framework distills the prominent structural features, dominant edges, medium-scale details, curved discontinued edges, silhouette, dendritic structures and curved boundaries and suppresses the superfluous details like noise, texture, irregular gradients, small-scale details and block artifacts. This framework effectively enhanced the significant image properties such as color, contrast, edge strength and sharpness at every stage based on the obtained statistical features availability information and the predefined conditions. This leads to enhancement of quality assessment features such as PSNR, SSIM and suppressing the image complexity and noise. It considers image and object space information to produce abstraction and stylization, thereby identifying emphasized elements of the structure using Harris feature detector algorithm. The proposed framework effectively preserves the structural features in the foreground of an image by comprehensively integrating the sequence of NPR image filters through rigorous experimental analysis simultaneously diminishing the background content of an image. Implementation of the proposed work is carried out in MATLAB 2018 with high-performance computer of 6.6 teraflop/s computing environment and Nvidia Tesla P100 GPU. The proposed framework evaluates every stage output with various subjective matters and quality assessment techniques with various statistical essences. By this manner, contextual features in an image have been identified and well preserved. Effectiveness of the proposed work has been validated by conducting the experiments taking David Mould dataset and Flickr images as references and comparing the obtained results with similar contemporary work cited in the literature. In addition, user’s visual feedback and the standard quality assessment techniques were also used to evaluate the work. Finally, this work lists out the structures preserving applications, constraints, framework implementation challenges and future work in the fields of image abstraction and stylization.

Underexposed heterogeneous complex-background and graphical embossing text documents are treated using proposed preprocessing image-abstraction framework that can deliver the effective structure preserved abstracted output by manipulating visual-features from input images. Reading of the text character in such images is extremely poor; hence, the framework effectively boosted the significant image properties and quality features at every stage. Work effectively preserves the foreground structure of an image by comprehensively integrating the sequence of NPR filters and diminishes the background content of an image, and in this way, the framework contributes to separation of foreground text from image background. Effectiveness of the proposed work has been validated by conducting the trials on the selected dataset. In addition, user's visual-feedback and image quality assessment techniques were also used to evaluate the framework. Based on the obtained abstraction output, this work extracts text-character by wisely utilizing traditional image processing techniques with an average accuracy of 98.91%.