Browsing by Author "Sequeira, A."

Now showing 1 - 2 of 2

A new deep learning architecture for dehazing of aerial remote sensing images
(Springer, 2022) Kalra, A.; Sequeira, A.; Manjunath, A.; Lal, S.; Raghavendra, R.
A major problem in most aerial remote image processing applications is the presence of haze in images. It is a phenomenon by which particles in the atmosphere disperse light, thus altering the quality of the overall image. This can be detrimental to the performance of vision-based algorithms such as those concerned with object detection. There have been numerous attempts using traditional image processing techniques as well as using deep learning approaches to eliminate this haze. In most cases, models tend to make assumptions on the nature of haze that are rarely true in reality. In this paper, we propose an end-to-end deep learning architecture that can dehaze aerial remote sensing images efficiently with minimal deviation from the ground truth. Many of the assumptions made in other models are eliminated and the relationship between hazed and dehazed images is directly computed. The proposed model is based on the observation that identifying structural and statistical portions separately from an image and using those features to reconstruct the image can give a realistic dehazed image. It also makes use of information exposed by different color spaces to achieve this using lesser computation. The experimental quantitative and qualitative results of the proposed architecture are compared with recent benchmark dehaze models on NYU hazy dataset and real-world hazy images. Experimental results yield that the proposed architecture outperforms benchmark models on test aerial remote sensing images. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
Scalable and Efficient orchestration of machine learning workloads on DSPs with multi-level memory architecture
(Society for Imaging Science and Technology, 2024) Sequeira, A.; Sam, F.; Jain, A.; Swami, P.
Deep learning has enabled rapid advancements in the field of image processing. Learning based approaches have achieved stunning success over their traditional signal processing-based counterparts for a variety of applications such as object detection, semantic segmentation etc. This has resulted in the parallel development of hardware architectures capable of optimizing the inferencing of deep learning algorithms in real time. Embedded devices tend to have hard constraints on internal memory space and must rely on larger (but relatively very slow) DDR memory to store vast data generated while processing the deep learning algorithms. Thus, associated systems have to be evolved to make use of the optimized hardware balancing compute times with data operations. We propose such a generalized framework that can, given a set of compute elements and memory arrangement, devise an efficient method for processing of multidimensional data to optimize inference time of deep learning algorithms for vision applications. Â© 2024, Society for Imaging Science and Technology.