scispace - formally typeset
Search or ask a question
Author

Raghav Sharma

Bio: Raghav Sharma is an academic researcher from Indian Institute of Technology Mandi. The author has contributed to research in topics: Object detection & Computer science. The author has an hindex of 1, co-authored 3 publications receiving 1 citations.

Papers
More filters
Book ChapterDOI
03 Sep 2019
TL;DR: A novel technique based on a modified SSD architecture OIR-SSD has proposed for real-time object detection on aerial images attaining high mean Average Precision (mAP).
Abstract: Aerial images usually are huge (around 2K resolution). Such high-resolution images contain thousands of small objects, and detecting all of them is a very challenging problem. The complexity of detection and classification in real-time is much higher than the usual images (<1K with high Object to Image Ratio OIR). Deep learning has many algorithms for object detection, but they are not designed for handling aerial images, and these algorithms are often sub-optimal for small scale object detection and their precise localization. In this work, a novel technique based on a modified SSD architecture OIR-SSD has proposed for real-time object detection on aerial images attaining high mean Average Precision (mAP). OIR-SSD has two approaches. The approach-I proposed for higher mAP, whereas the approach-II proposed to achieve real-time object detection. The approach-I has improved mAP from 0.72 to 0.92 (28% improvement) on Stanford data-set while from 0.04 to 0.44 (1100% improvement) on Visedrone2018 at 4 Frames Per Second (FPS) whereas the approach-II has improved mAP from 0.72 to 0.82 at 42 FPS.

1 citations

Journal ArticleDOI
TL;DR: A hybrid approach using a machine learning technique called eigenfaces, along with vanilla neural networks is discussed, which proved to be more promising and efficient than its counters.
Abstract: Coronavirus has become one of the most deadly pandemics in 2021. Starting in 2019, this virus is now a significant medical issue all over the world. It is spreading extensively because of its modes of transmission. The virus spreads directly, indirectly, or through close contact with infected people. It is proclaimed that people should wear a mask in public areas as a counteraction measure, as it helps in suppressing transmission. A portion of the spaces, where the virus has broadly fanned out, is because of inappropriate wearing of facial cover. In crowded areas, keeping a check on facial masks manually is difficult. To automate this process, an effective and robust face mask detector is required. This paper discusses a hybrid approach using a machine learning technique called eigenfaces, along with vanilla neural networks. The accuracy was compared for three different values of principal components. The test accuracy achieved was 0.87 for 64 components, 0.987 for 512 components, and 0.989 for 1,000 components. Hence, this approach proved to be more promising and efficient than its counters.

1 citations

Book ChapterDOI
01 Jan 2021
TL;DR: In this article, a real-time object detector architecture, sliding window detector (SWD) based on a sliding window technique, has been proposed, which can execute on a low-compute device.
Abstract: In recent years, deep learning-based object detection has been researched hot spot due to its powerful learning ability in dealing with occlusion, scale transformation, and background switches, etc. There are many state-of-art object detectors like SSD, SSH, YOLO, and RCNN that have been invented in recent years. These architectures are highly complex, work on deep learning framework, requiring high computing power, which restricts their practical adaptability for low-cost applications and low-computes devices. In this work, a novel real-time object detector architecture, sliding window detector (SWD) based on a sliding window technique, has been proposed. SWD works on a deep learning framework and can execute on a low-compute device. In the proposed SWD architecture, the classifier network is optimized. The fully connected layer of the classifier trained on N classes is replaced by a convolutional layer, which generates N heat-maps. These heat-maps are used to localize and classify the object. SWD simulation on an Intel i5 CPU with 20 FPS shown mAP 0.85 for PKLOT data-set.

1 citations

Journal ArticleDOI
TL;DR: In this article , a real-time, precise, and efficient approach based on machine learning to recognize maladies existing in a tomato plant by revealing from its leaves was presented, where a convolution neural network-based algorithm was used for the identification of maladies in tomato plant leaf.
Abstract: Tomato crops are progressively mainstream, as a result of their high nutrition power which is available such as beta-carotene and Vitamin C, E. Lack of care for such crops results in causing serious ailments on plants and it is very important for identifying disease in plants for an efficient crop yield. Maladies on the plant causes depletion in both the quantity and quality of the crop. This paper possesses a real-time, precise, and efficient approach based on machine learning to recognize maladies existing in a tomato plant by revealing from its leaves. The model for disease detection was trained on plant leaves which were taken from PlantVillage Dataset and a convolution neural network-based algorithm was used for the identification of maladies in tomato plant leaf. The outcome shows that the presented approach is effective in detecting disease of tomato plant leaves as well as it could be globalized.

1 citations

Proceedings ArticleDOI
04 Sep 2022
TL;DR: This paper deploys two novel bi-directional encoder-based systems, viz., BioBERT and RoberTa to identify named entities in the biomedical text and obtains a significant improvement in F-score by RoBERTa over BioberT and a comparative study on training loss attained with ADAM and LAMB optimizers.
Abstract: The recent advancements in medical science have caused a considerable acceleration in the rate at which new information is being published. The MEDLINE database is growing at 500,000 new citations each year. As a result of this exponential increase, it is not easy to manually keep up with this increasing swell of information. Thus, there is a need for automatic information extraction systems to retrieve and organize information in the biomedical domain. Biomedical Named Entity Recognition is one such fundamental information extraction task, leading to significant information management goals in the biomedical domain. Due to the complex vocabulary (e.g., mRNA) and free nomenclature (e.g., IL2), identifying named entities in the biomedical domain is more challenging than any other domain, hence requires special attention. In this paper, we deploy two novel bi-directional encoder-based systems, viz., BioBERT and RoBERTa to identify named entities in the biomedical text. Due to the domain-specific training of BioBERT, it gives reasonably good performance for the NER task in the biomedical domain. However, the structure of RoBERTa makes it more suitable for the task. We obtain a significant improvement in F-score by RoBERTa over BioBERT. In addition, we present a comparative study on training loss attained with ADAM and LAMB optimizers.

Cited by
More filters
Journal ArticleDOI
TL;DR: In this article , the authors surveyed and compared publicly available image datasets specifically crafted to test computer vision-based methods for parking lot management approaches and consequently presented a systematic and comprehensive review of existing works that employ such datasets.
Abstract: Computer vision-based parking lot management methods have been extensively researched upon owing to their flexibility and cost-effectiveness. To evaluate such methods authors often employ publicly available parking lot image datasets. In this study, we surveyed and compared robust publicly available image datasets specifically crafted to test computer vision-based methods for parking lot management approaches and consequently present a systematic and comprehensive review of existing works that employ such datasets. The literature review identified relevant gaps that require further research, such as the requirement of dataset-independent approaches and methods suitable for autonomous detection of position of parking spaces. In addition, we have noticed that several important factors such as the presence of the same cars across consecutive images, have been neglected in most studies, thereby rendering unrealistic assessment protocols. Furthermore, the analysis of the datasets also revealed that certain features that should be present when developing new benchmarks, such as the availability of video sequences and images taken in more diverse conditions, including nighttime and snow, have not been incorporated.

13 citations

Book ChapterDOI
01 Jan 2021
TL;DR: In this article, a real-time object detector architecture, sliding window detector (SWD) based on a sliding window technique, has been proposed, which can execute on a low-compute device.
Abstract: In recent years, deep learning-based object detection has been researched hot spot due to its powerful learning ability in dealing with occlusion, scale transformation, and background switches, etc. There are many state-of-art object detectors like SSD, SSH, YOLO, and RCNN that have been invented in recent years. These architectures are highly complex, work on deep learning framework, requiring high computing power, which restricts their practical adaptability for low-cost applications and low-computes devices. In this work, a novel real-time object detector architecture, sliding window detector (SWD) based on a sliding window technique, has been proposed. SWD works on a deep learning framework and can execute on a low-compute device. In the proposed SWD architecture, the classifier network is optimized. The fully connected layer of the classifier trained on N classes is replaced by a convolutional layer, which generates N heat-maps. These heat-maps are used to localize and classify the object. SWD simulation on an Intel i5 CPU with 20 FPS shown mAP 0.85 for PKLOT data-set.

1 citations

Proceedings ArticleDOI
23 Dec 2022
TL;DR: In this article , a visual geometry group (VGG) model based on CNN is employed to detect tomato plant leaf illnesses more quickly, achieving an accuracy of 97.67% and 87.67%.
Abstract: Plant diseases that may seriously impact agriculture are often discovered with the naked eye, albeit this can take more time and increase the likelihood of a false positive. This issue may be resolved and the chance of decreased plant output is decreased with early discovery. The aim of this experimental research is to deploy intelligence, which can be effectively used for picture classification utilising numerous convolutional neural network (CNN) manners, to automatically identify tomato plant leaf illnesses more quickly. For better performance measurement, the Visual Geometry Group (VGG) model, which is based on CNN, is employed. To diagnose illnesses, this research concludes to categorise photos using VGG-19 transfer learning architectures with various optimizers. In the experimental comparative research, an accuracy of 97.67% and 87.67% was achieved as training and testing with nadam optimizer.
TL;DR: A hybrid approach using a machine learning technique called eigenfaces, along with vanilla neural networks is discussed, which proved to be more promising and efficient than its counters.
Abstract: Received Apr 11, 2022 Revised Sep 16, 2022 Accepted Sep 30, 2022 Indigital circuits, energy reduction is the most important parameter in the design of handy and battery-operated devices. Flipflop is an important component in any digital system. By improving the performance of flip-flop, complete system performance is better. This paper addresses the design of D flip-flop using direct current diode-based positive feedback adiabatic logic (DC-DB PFAL) at various frequencies at 45nm technology node. Further, the layout for the proposed design is also presented. The performance analysis is carried out for delay, power dissipation, power delay product and transistor count. Circuit simulation is done by using cadence virtuoso tool at 10 MHz and 100 MHz frequencies. The results show an improvement in power dissipation of 18% with less transistors count compared to exiting designs in the literature.