SJSurekha Janrao

Surekha Janrao

*This is an algorithmically generated author record based on publicly available information.

Surekha Janrao has a total of 32 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Improving the quality of images using Wasserstein generative adversarial networks for image restoration

    by Aruna Pavate, Surekha Janrao, Rohini Patil, Maganti Venkatesh, Shudhodhan Bokefode, Yunfei Li, Ubaldo Comite

    In the present digital age, it is crucial to preserve personal memories and historical photographs in their original form, and this is made possible through image restoration. This paper presents a dynamic multi-scale Wasserstein generative adversarial network with gradient penalty (WGAN GP) framework that combines colorization and image denoising, addressing the limitations of distinct restoration models that denoise and colorize images in parallel. The proposed system adapts to hierarchical image features, stabilizes training, and enhances fine-grained texture reconstruction. The model is trained on CelebA, Places365, and ImageNet datasets. The need for repeated retraining is required, and there are still no guarantees of robustness under various degradations such as fading, saturation loss, and sensor noise. The results show peak signal-to-noise ratio (PSNR) of 24.5 dB and structural similarity index measure (SSIM) of 0.74, outperforming Pix2Pix, CycleGAN, denoising generative adversarial network (D‑GAN), and enhanced super‑resolution generative adversarial network (ESRGAN) in efficiency and robustness. In contrast to previous GAN-based restoration methods that treat denoising and colorization as separate problems, the presented multi-scale WGAN-GP applied a generator-discriminator model, resulting in less training redundancy and similar SSIM results while using ~55-65% less number of training epochs than ESRGAN and DeblurGAN. In the future, the model will integrate attention and transformer-based refinement to enhance detail recovery and perceptual realism further.

  • Open Access

    Deep learning approaches for Braille detection and classification: comparative analysis

    by Surekha Janrao, Tavion Fernandes, Ojas Golatkar, Swaraj Dusane

    This study proposes a hybrid approach to Braille translation leveraging the strengths of both YOLO for object detection and multitude of classification models such as ResNet, and ResNet for accurate Braille character classification from images. Upon comparing numerous models on various performance metrics, ResNet and DenseNet outperformed other models, exhibiting high accuracy (0.9487 and 0.9647 respectively) and F1-scores (0.9481 and 0.9666) due to their deep, densely connected architectures adept at capturing intricate Braille patterns. CNNs with pooling showed balanced results, while MobileNetV2's lightweight design limited complex classification. ResNeXt's multi-path learning achieved respectable performance but lagged behind ResNet and DenseNet. In the future the results from our study could be further explored on contracted Braille recognition, be adapted to various Braille codes, and optimized for mobile devices, for real time Braille detection and translation on smartphones.

  • Open Access

    Efficient model for cotton plant health monitoring via YOLO-based disease prediction

    by Aruna Pavate, Swetta Kukreja, Surekha Janrao, Sandip Bankar, Rohini Patil, Vijaykumar Bidve

    Protecting plants from diseases involves recognizing the symptoms and identifying practical, safe, and reasonable treatment methods. Holistic approaches based on particular times or seasons can reduce plant resistance and minimize tedious work. Technological advancements have led to the development of microscopic examinations and computational methods using machine learning techniques to detect diseases automatically and quickly using leaf images. This study builds the prediction model using EfficientNet and YOLO neural network architectures from computer vision. The development of a model that assists farmers in identifying cotton disease so that they use pesticides that may treat it further utilizes this concept. In the physical world, the input is accepted from many different sources, so observing the model’s output is necessary. This work concentrates on model response to the inputs from physical devices, and analysis shows that the monitoring varies the results. A novel convolutional neural network (CNN) based on the EfficientNet architectures and variations of YOLO architectures is used to classify and identify the objects in cotton leaf. The EfficientNetB4 yielded 100% accuracy for healthy leaf and powdery mild leaf classes, and YOLO v4 version with 96%, 98.3%, 99.2%, and 0.70 for precision, recall, mAP@0.5, mAP120.5:095 respectively. These results indicate that consequences vary in real-time per environmental parameters such as light effect and devices, and analysis shows that monitoring affects the results.


  • Open Access

    A Framework for Prediction of Type II Diabetes through Ensemble Stacking Model

    by Rohini Patil, Anant Patil, Surekha Janrao, Sandip Bankar, Kamal Shah

    In order to prevent long term complications of diabetes its early diagnosis is crucial. With Increasing advances in Artifical Intelligence (AI) and Machine Learning(ML) researchers are increasingly focusing on using them for early diagnosis of diseases.AI and ML has significant potential for early prediction of type 2 diabetes. In this article we have described a ML based framework for prediction of type 2 diabetes -Improved Ensemble Learning with Dimensionality Reduction Model (IELDR) and discussed its result. An IELDR algorithm is an Auto encoder-based feature extraction method with ensemble learning. The experiments were carried out using the LS_diabetes dataset. LS_diabetes dataset containing 374 records with 35 features related to lifestyle and stress. Accuracy, precision, specificity, sensitivity, f1 score, roc and Mathew correlation coefficient (MCC) were measured. After this results were tested and validated using Diabetes_2019 dataset and PIMA diabetes dataset. The IELDR model showed results in terms accuracy, precision, specificity, sensitivity, f1 score, roc and Mathew correlation coefficient (MCC) of 98.67%, 95.24%, 100%, 98.18%, 97.56%, 99.09% and 0.97 respectively. In comparison with PIMA diabetes dataset, LS_diabetes dataset showed an accuracy, precision, sensitivity, specificity, f1-score,roc and mcc value by 17.96%,13.15% 40.22%,5.59%,28.38%,22.09% and 0.4 respectively. The IELDR model achieved the best result on the LS_diabetes dataset showed an accuracy, sensitivity, roc and mcc value improved by 1.82%, 1.58%, 3.01%and 0.04 % compared to the Diabetes_2019 dataset .This proposed IELDR system predicts the risk of type 2 diabetes in a healthy person based on the person’s current lifestyle pattern. This system can be helpful for early prediction of type2 diabetes.


  • Open Access

    Conglomerate Crop Recommendation by using Multi-label Learning via Ensemble Supervised Clustering Techniques

    by Surekha Janrao, Kamal Shah, Aruna Pavate, Rohini Patil, Sandip Bankar, Anil Vasoya

    Existing crop recommender related to either binary classification or multiclass classification. This paper presents conglomerate crop recommendations which consist of a number of different and distinct crops that are grouped together. In this work we focus on transferring knowledge from single label output prediction to multiple label predicted output for a given input data instances. We proposed ESCT algorithm i.e. Ensemble Supervised Clustering Techniques in our research work. ESCT provides a combined approach of conventional clustering and enhanced supervised clustering methodology to optimize the conglomerate recommendation. We are focusing on K-mean clustering for conventional approach and ICCC i.e. Inter cluster correlation coefficient to achieve enhancement in supervised clustering. In conventional K-mean clustering there is a big challenge on how to optimize the k-value of clustering which directly affects the convergence of the clusters. To resolve this problem, we mainly apply function approximation on K-Value which provides us with better clustering and fast convergence. Existing methods for inter-clustering do not adequately address one of the key challenges i.e. exploiting correlations between labels and that is achieved by ICCC algorithm. This model provides learning and prediction of unknown observation by using Back propagation MLL algorithm which provides improved performance.


  • Open Access

    A Novel Approach for Iris Recognition System Using Genetic Algorithm

    by Jayesh Mohanrao Sarwade, Sandip Bankar, Surekha Janrao, Kishor Sakure, Rohini Patil, Shudhodhan Bokefode, Nilesh Kulal

    In a brand-new era, with chaotic scenario that exists within the world, people are undermined with diverse psychological assaults. There have been numerous sensible approaches on the way to understand and lessen those attacks. Bioscrypt developments have verified to be one of the beneficial approaches for intercepting these troubles. Identifying recognition through human iris organ is said as one of the well-known biometric strategies because of its reliability and higher accurate return in comparison to different developments. Reviewing beyond literatures, terrible imaging condition, low flexibility of version, and small length iris image dataset are the constraints desiring solutions. Among these kinds of developments, the iris popularity structures are suitable gear for the human identification. Iris popularity has been an energetic studies location for the duration of previous couple of decades, due to its extensive packages in the areas, from airports to native land protection border protection. In the past, various functions and methods for iris recognition have been presented. Despite of the very fact that there are many approaches published in this field, there are still liberal amount of problems in this methodology like tedious and computational intricacy. We suggest an all-encompassing deep learning architecture for iris recognition supported by a genetic algorithm and a Wavelet Transformation, which may jointly learn the feature representation and perform recognition to realize high efficiency. With just a few training photos from each class, we train our model on a well-known iris recognition dataset and demonstrate improvements over prior methods. We think that this architecture can be frequently employed for various biometric recognition jobs, assisting in the development of a more scalable and precise system. The exploratory aftereffects of the proposed technique uncover that the strategy is effective inside the iris acknowledgment.


  • Open Access

    Return on investment framework for profitable crop recommendation system by using optimized multilayer perceptron regressor

    by Surekha Janrao, Deven Shah

    Return on investment (ROI) plays very important role as a financial dimension in the agriculture sector. Many government agencies like Indian space research organization (ISRO), Indian council of agricultural research (ICAR), and Nitiayog are working on different agriculture projects to improve profitability and sustainability. This paper presents ROI framework to recommend more profitable crop to the farmers as per the current market price and demand which is missing in the existing crop recommendation system. Crop price prediction (CPP) and crop yield prediction (CYP) system are integrated in the ROI framework to predict more demandable crop to yield. This framework is designed by applying data analysis to provide regression statistics which further helps for model selection and improve the performance also. Optimized multilayer perceptron regressor algorithm has been evaluated through experimental results and it has been observed that it gives better performance as compared to other existing regression techniques.