Recent :
[paper] [Presentation] Bird Species Classification from an Image Using VGG-16 Network. In this paper, the main objective is to classify Bangladeshi birds according to their own species using several machine learning algorithms through transfer learning. The dataset for this classification is one that I collected manually and consists only of bird species that are found in Bangladesh. This was done because there is no collection of local bird data in Bangladesh. We used the VGG-16 network as our model to extract the features from bird images. In order to perform the classification, we used several Algorithms. However, compared to other classification methods such as Random Forest and K-Nearest Neighbor (KNN), Support Vector Machine (SVM) gave us the maximum accuracy of 89%.
[Paper Under Review] Detecting Fake Bengali Name From Social Network. In this paper, we explore natural language processing techniques for the detection of a fake name in the context of Bangladesh. To extract name features we use Term Frequency-Inverse Document Frequency and N-grams of letters. After getting vector features we feed those models into the ML pipeline. The main techniques are supervised learning pipelines to classify names whose names do not represent real Bengali names. We ran different Machine Learning algorithms such as Naïve Bayes, Random Forest (RF), and Support Vector Machine (SVM).
[Poster][Video][Paper Under Review] Real-Time Traffic Detection And Management from an Image using Machine Learning Approach. We introduce a traffic system that is more robust and minimize waiting time. The system intends to identify high traffic and low traffic in order to control traffic on each side of the road with the help of a machine learning model. In this paper, we use a transfer learning technique where Pre-trained CNN networks are loaded to extract features from the image. The system will take a video feed of the roads and feed it to a machine learning model. The overall system will consist of a server and a client-side. The machine learning model is made up of a VGG-16 network for the feature extraction and uses SVM to perform the classification.
[Demo]Bangla OCR: This research is a part of a project called "Bangladeshi Car License Plate Detection and Recognition".in this research we use 10k Bangla numbers and the letter for recognition. we use Mobilenet and our own neural network architecture for classification. we achieve 99% accuracy for Bangla digits and letters.