Deep Neural Network for Recognition of Enlarged Mathematical Corpus
Introduced an approach to recognize symbols in the hybrid dataset of handwritten mathematical expressions.
A Novel Approach to Detect Mathematical Expressions: Recognition Based Convex Hulls
A new approach is introduced to perform segmentation on the handwritten symbols to minimize memory occupancy
DenseNet-based Attention Network to recognize handwritten mathematical expressions
The attention-based model recognizes the segmented expression with the ExpRate of 57.4% on the custom dataset.
Applying deep learning in Mars exploration: a neural network-based study to classify Martian terrain features
We proposed a DNN-based procedure to recognize and classify the features of Martian terrain with an accuracy of 94.8%.
End to End Deep Neural Network: An approach to clean noisy documents
In our study, we introduce an end-to-end deep learning neural network approach to clean the noisy document with an accuracy of 66.3%.
Accepted/Presented/To-be-published Papers
Impact of Varying Strokes on Recognition Rate: A Case Study on Handwritten Mathematical Expressions [Accepted in
IJCDS journal]
A detailed and comprehensive study of the impact of varying stroke width on the recognition rate of the characters by a deep learning model.
Deep Learning Based Method To Detect Diseases In Leaves Of Cassava Plant [Presented at ICDABI'22]
In this article, we discuss the deep learning-based method to detect and classify the disease by images of cassava plant leaves. We achieved an accuracy of 84.3%.
Railway Track Defect Detection using Transfer Learning With EfficientNetB3 [Presented at ICDABI'22]
By using the transfer learning method and EfficientNetB3, we built a customized DNN network to detect defects in railway tracks with an accuracy of 93.55%.
A Deep Learning Based Efficient Prediction Model for Early Stage Detection of Cervical Spine Problems [Accepted to be presented at SRC'22 conference]
By application of a deep neural network, we proposed an automated methodology to predict various problems associated with
the cervical spine with an accuracy of ≈99%.
XceptionNet-based Transfer Learning Model for Weather Image Recognition [Accepted to be presented at SRC'22 conference]
We employed a customized DNN model based on the XceptionNet transfer learning method(model) to predict the type of
weather in the scene with an accuracy of 88.2% without fine-tuning the base model.
Classification of Common Insects using Transfer Learning and MobileNetV2 [Accepted to be presented at SRC'22 conference]
Using the latest resource-efficient DNN model named MobileNetV2, we built our own custom neural network to perform the
recognition class of insects. The model achieved an F1 score of 85.3%.
Automating Reverse Image Search: An Approach to Detect Discrepancy in Images Using Selenium and Python. [Accepted to be presented at SRC'22 conference]
We developed a GUI app with the feature to perform a reverse image search to annotate whether or not images are
anachronistic.
The program is tested on single and multi-thread environments with various input modes. The time taken in each case and mode
is far less than the traditional method.
All so Perfect Model: MobileNetV3-based classification agricultural crop images. [Accepted to be presented at SRC'22 conference]
To automate the damage estimation during natural calamities, we developed a TL-based DNN model to recognize and classify the
type of crops present in the image with an accuracy of ≈100%.