<p><em>AdaBoost along with HaarCascades have been well received for its accuracy and performance in primarily Facial Recognition applications. However, they are known to perform poorly with objects which have a different rotational orientation or for objects whose shapes are largely variant . In this paper, we apply Adaptive Cascading technique to a specific dermatological application of detecting red acne which are largely shaped variant outgrowths on the skin and to identify its suitability in the detection of acne. Based on the outcome it would be declared if Viola-Jones based Adaptive Boosting is well suited for dermatological processing of skin diseases.</em></p>
Introduction: Bow Legs and Knock Knees are quite common in growing children, which usually affect the lower portions of the body, however such disorders usually do not have any pathological significance. In this paper, we investigate a method using deep learning to correctly draw a boundary between a physiologically normal knee and a genu valgum. Objective: To draw a decision boundary between what is classified as Normal and what is "Abnormal" i.e. a knee exhibiting features of Knock knees which is Genu Valgum by using AI and ML tools. Methods: For this study the Adam Gradient descent was used which is a combination of AdaGrad and RMSProp. There is also an implementation of grid search for "self-selection" of parameters by the neural network which is the unique point that most existing ML algorithms on account of self-learning capability much like un-supervised learning but limited to parameter selection. In the second part, we try to investigate the outcome using X-ray version of the disorder and try to compare if the result is truthful in accordance to the patient's case. Results: The two types of Knees had been correctly classified up to an accuracy of 89% to 90% (by using normal to normal) which is really good for most physicians or sports instructors to use as an initial screening tool for most athletes/patients. However, the second part shows interesting results with an accuracy of 60% (X-ray to Normal).
Falls affect seniors' quality of life, and therefore fall detection and prevention are paramount for the health and safety of aging seniors. Current deep learning-based fall detection methods perform well when a large amount of training data is available. As obtaining fall data from seniors is extremely difficult, training deep learning models is a challenge, and therefore, a few-shot Siamese network is considered in this thesis. A shallow 1 × 1 convolutional neural network for Siamese and Triplet networks is proposed in this work. A deeper architecture-based on the Inception and Densenet networks is also considered to improve the fall detection performance. The performances of the proposed few-shot Siamese architectures and Triplet networks are investigated using signals obtained from a wearable sensor. The proposed learning models outperform the traditional deep learning networks, while Siamese architectures also demonstrate generalizability by classifying unseen classes of falls and falls from different sensing modalities. First and foremost I am extremely grateful to my supervisor, Dr. Sreeraman Rajan, for his awesome advice, continuous support, and patience during the course of my masters program. His plentiful experience have encouraged me through out the time of my research and daily life. I thank the examination committee members for their time and efforts to evaluate my work and provide valuable feedback. I am extremely grateful to Dr. Jila Hosseinkhani for meticulously proof-reading my thesis and providing appropriate feedback. I would like to thank the Department of Systems and Computer Engineering, Carleton University, for providing me the opportunity and on-demand support for the completion of my program. I also thank the Natural Science and Engineering Council of Canada (NSERC) for partially funding the work. I also acknowledge the financial support that I got for a term from Department of National Defence's (DND)-IDEAS-Micronet initiative. These financial supports have made my study and life in Canada a smooth experience through out the ups and downs of this pandemic-ridden world. Finally, I would like to express my gratitude to my parents. Without their financial support, tremendous understanding and encouragement in the past few months, it would be impossible for me to complete my program.
In recent years, Attention transformers have proven to be instrumental in Natural Language Processing (NLP) based tasks like sentence classification, and language translation. However, their application has been recently extended to large-scale object recognition tasks. In this work, Vision Transformer with attention has been investigated for the detection of human falls and ADLs (Activities of Daily Living) from time series-based signals. The Vision Transformer model has been trained and validated using the accelerometer signals of waist-worn Inertial Measurement Unit (IMU) sensors obtained from the IMU Falls dataset. The model is also trained and validated on the popularly used SiSFall dataset. The model is also investigated by independently training on 3 different cases of patch size and attention heads. It is observed that a larger patch size has resulted in significant performance deterioration. Additionally, smaller patch size took longer to train and was computationally expensive. The model performed (best-case) with an Accuracy (%) of 99.9 ± 0.1 and a True Positive Rate (%) of 99.9 ± 0.1 on the SFU-IMU dataset and with an Accuracy (%) of 99.8 ± 0.25 and a True Positive Rate (%) of 99.87 ± 0.3 on the SISFALL dataset. Overall, the results show that Transformers are highly robust in the detection of human falls and non-falls/ADL subject to the appropriate patch size.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.