C. Gururaj's research while affiliated with Alpha College of Engineering Bangalore and other places

The technique of segmenting a picture involves breaking it up into its component parts, such as boundaries and regions. Image edges are found via Edge detection algorithms. Edges of the picture are detected when the pixels’ intensities abruptly shift. Sobel, Prewitt, and Robert are three well-known gradient-based edge detection algorithms that operate on the 3-by-3 and 2-by-2 kernels in the X and Y axis. Python and Verilog HDL for synthesis on the Artix 7-based Basys 3 dev kit are used to perform in two separate phases. The image data is first converted to grayscale for the Python implementation before being passed through the gradient-based edge detection (Sobel, Prewitt, and Robert) operator. The edges of the entire image are then determined using an edge detection technique. The resultant picture is then examined using various input parameters including PSNR, MSE. The input grayscale image is passed via line buffers, which store the pixels for subsequent processing by the operators, in the second stage, which is Verilog implementation. Output buffer is then filled based on the output from the convolution operator. The kernels of 3 × 3 or 2 × 2 for the convolution are pre-defined based on the operator.KeywordsEdge detectionFPGASobel edgePrewitt edgeRobert edgeImage segmentation

Thyroid nodule is a common disease on a global scale. It is characterized by an abnormal growth of the thyroid tissue. Thyroid nodules are divided into two types: benign and malignant. To ensure effective clinical care, an accurate identification of thyroid nodules is required. One of the most used imaging techniques for assessing and evaluating thyroid nodules is ultrasound. It performs well when it comes to distinguishing between benign and malignant thyroid nodules. But ultrasound diagnosis is not simple and is highly dependent on radiologist experience. Radiologists sometimes may not notice minor elements of an ultrasound image leading to an incorrect diagnosis. After performing a comparative study of several deep learning-based models implemented with different classification algorithms on an open-source data set, it has been found that ResNet101v2 gave the best accuracy (~96%), F1 score (0.957), sensitivity (0.917), etc. A simple and easy-to-use graphical user interface (GUI) has also been implemented.

During Covid-19, many supply chains were disrupted. Supply chain resilience can be improved by developing business continuity capabilities using artificial intelligence (AI). This research examines how companies use artificial intelligence (AI) and looks at ways that AI can improve supply chain resilience by increasing visibility, reducing risk, and improving sourcing and distribution. Early detection of SARS-CoV-2 (2019-nCoV), which is caused by the lethal virus SARS-Cov-2 (Severe Acute Respiratory Syndrome Corona virus), has become critical especially as the epidemic spreads. X-rays and computed tomography scans are examples of medical imaging that can help in diagnosis. CT scans are preferable over RT-PCR tests because of their inaccuracy. In this era of fast technological growth, using artificial intelligence methodologies to construct models with a higher performance volume and better accuracy predictions is a huge step forward. Medical image analysis incorporating image processing and computer vision techniques were used to analyse the chest Radiographs and train the models. The accuracy and significant amount of data collection and prediction supply chain for efficient detection of COVID-19 utilising Artificial Intelligence techniques are described in this study. Models are built using data obtained by local CT scan centres. The data can be reviewed from time to time in coordination with CT scan centres. The application will provide accurate predictions so this has a significant impact on the tool's market worth. For the post-COVID-19 period, many firms are hastening the creation of management plans with supply chain transformation in mind. In this pandemic, but even so, the market will be even narrower, so without using a decentralised governance framework with an imbalanced structure among various markets, it really should be moved to a centralised management strategy that combines advantage of the existing strength of a blocked setup, with almost as much vicinity to the manufacturing countries and regions. In a stronger emphasis Supply Chain Management, value management, in value analysis, plays a significant role. Real-time raw data of chest CT scans from hospitals were considered and used it to train the model after pre-processing it. In a chest CT scan, multiple perspectives and organs are focused, but the work solely used the axial perspective of the lungs to prepare the dataset. Around 1900 photos of each COVID and Normal are included in the dataset. The data was pre-processed with a range filter for noise reduction, cropping, data augmentation, and other minor operations such as adjusting the image brightness and sharpness. Deep Learning algorithms are trained using this pre-processed data. VGG16, ResNet101, Inception v2, DesneNet169, and Mobile net are implementations of deep learning algorithms that is developed. The same dataset was used to train the above models, however because each model has a distinct architecture, the accuracy of the models varies slightly. The test dataset for all of the models includes 300 images in each class, and the findings demonstrate that DenseNet169 has the best accuracy among the models, while ResNet101 has the poorest. Furthermore, the medical image analysis of Covid-19 by several models aids in the selection of the most accurate model for COVID-19 predictions from CT scans. A windows application for the prediction of COVID has been developed where the user will upload the CT scan image and has an option to specify model and get the prediction. If tested positive, user can also view the infected area in the image. This application uses DenseNet169 as default model, in case user does not specify the model, as it performs the best. The user can view the COVID protocols and related queries from WHO website using a query button. The availability of CT scans and other commodities for this application varies as a result of the pandemic, which has an impact on the global supply chain and the market price of this tool.

Nowadays memes have become a way in which people express their ideas on social media. These memes can convey various views including offensive ones. Memes can be intended for a personal attack, homophobic abuse, racial abuse, attack on minority etc. The memes are implicit and multi-modal in nature. Here we analyze the meme by categorizing them as offensive or not offensive and this becomes a binary classification problem. We propose a novel offensive meme classification using the transformer-based image encoder, BiLSTM for text with mean pooling as text encoder and a Feed-Forward Network as a classification head. The SwinT + BiLSTM has performed better when compared to the ViT + BiLSTM across all the dimensions. The performance of the models has improved significantly when the contextual embeddings from DistilBert replace the custom embeddings. We have achieved the highest recall of 0.631 by combining outputs of four models using the soft voting technique.