KOCHI:
Naval Research Board has sanctioned the research project entitled “Machine learning Models for underwater Image Enhancement and Content Analysis” with a total cost of Rs. 48,04,560/-.
Aim of the project is to Implement Machine Learning Models Based on Deep Neural Network Algorithms capable of image analytics, which can be employed as an assistive automated real-time monitoring system with Autonomous Underwater Vehicles (AUVs). The project is intended to be carried out in the Advanced Signal Processing and Instrumentation Research lab (ASPIRE) as well as Cochin University Centre for Ocean Electronics (CUCENTOL), Department of Electronics, Cochin University of Science and Technology (CUSAT)in collaboration with DRDO Laboratory in Cochin.
Underwater image is inundated by reduced visibility conditions making images with deprived colour variation and contrast. The emphasis of this project lies primarily in the area of image processing and analysis of processed images. The underwater scenario due to its high dynamic nature poses high challenges to the conventional image processing algorithms.
Learning model based approach would perform better compared to the handcrafted feature engineering and mathematical modelling based techniques. Large amount of video footage recorded in the underwater scenarios especially marine ecosystems, demands high efficient monitoring, which is an indispensable task for maintaining marine ecosystem and the strategical areas related to national safety.
Monitoring of video footage is still utilizing manual labour to a higher extend in many fields, which is highly inefficient and prone to human errors. Automated techniques and intelligent detection systems may help in this regard. The project aims at enhancing the underwater image captured with the help of machine learning models and then by analyzing the content of the image through specialized deep neural network architectures and algorithms.
The tasks may be broken down into objectives such as image enhancement modelling, object detection modelling, segmentation algorithms and classifying the object to come up with the predictions. Learning models learns from the data it processed while training, and therefore the dataset generation need to be performed with much caution. Ground truth data has to be captured through video footage and thereby annotating and labelling the images extracted from video to form the database for training the model. Transfer Learning may also be explored utilizing the specialized, highly efficient architectures to extract features to perform the image content analysis. Thus developed learning models can serve as the intelligent assistant system for ROVs and AUVs for quick decision making and real-time underwater navigation.
Dr.Supriya M. H., Professor & Head, Department of Electronics is the Principal Investigator of the project. Arun A. Balakrishnan and Mithun Haridas T. P., Assistant Professors, Department of Electronics, Cochin University of Science and Technology (CUSAT)are the co-investigators involved in the project.
