Projects | Academic

Chicken Behavior Analysis

Fri, 01 Apr 2022 00:00:00 +0000

I successfully completed a project focused on monitoring chicken behavior through the application of cutting-edge computer vision techniques. Leveraging the power of deep convolutional neural networks, I implemented an approach for chicken detection, to accurately identify and track individual chickens across consecutive video frames. This innovative framework enabled a comprehensive analysis of chicken behavior, providing valuable insights into their movements and interactions within the poultry. I won a prestigious prize for the development and application of this novel technology in the realm of poultry behavior monitoring. This accomplishment not only highlights the effectiveness of my work but also underscores the significance of utilizing AI technologies to enhance our understanding of animal behavior and welfare.

The application of artificial intelligence (AI) in the context of animal welfare has the potential to revolutionize the way we understand, monitor, and enhance the well-being of animals across various environments. One of the primary benefits lies in the ability of AI to provide continuous and real-time monitoring of animal behavior, health, and overall conditions. For instance, AI-powered systems can analyze data from sensors and cameras to detect signs of distress, illness, or abnormal behavior in livestock, enabling prompt intervention and veterinary care. Additionally, AI facilitates the development of predictive models that can anticipate potential issues or health concerns before they escalate. This proactive approach allows for preventive measures to be implemented, ensuring a higher standard of care for animals. The integration of machine learning algorithms can also aid in identifying patterns related to optimal living conditions, feeding schedules, and environmental preferences, contributing to the creation of more tailored and animal-centric practices.

Project GitHub page: https://github.com/TaherehZarratEhsan/Chicken-Behavior-Analysis

Violence Recognition in Video Sequences

Fri, 01 Apr 2022 00:00:00 +0000

Throughout my research endeavors, I have developed a robust foundation in computer vision, machine learning, and deep learning. This expertise is reflected in seven published papers focusing on violence detection to date. In the initial stages, I designed handcrafted features based on motion patterns to discriminate between violent and normal actions. As my research progressed, I delved into deep learning techniques such as CNNs and LSTMs.

However, the effectiveness of deep networks relies heavily on extensive datasets to automatically identify patterns and extract features from complex data. Faced with the constraints of a limited annotated dataset, challenges in this domain often arise in generalizing to Out-of-Distribution data. Acknowledging this hurdle, I redirected my efforts toward enhancing generalizability. Leveraging the available large-scale normal action recognition datasets, I pre-trained the model to incorporate valuable knowledge, enabling it to comprehensively learn normal patterns. By framing violence recognition as anomaly action recognition, I successfully detected abnormal human actions. To achieve this, I proposed various techniques, including representation learning using AE and generative modeling through GAN. Notably, prior to my research, there had been no comprehensive investigation into increasing generalization in violence recognition.

In comparison to state-of-the-art methodologies, my contributions have yielded substantial results, showcasing a remarkable 40% improvement in generalization accuracy across new real-world environments.

Breast Cancer Detection

Sat, 11 Dec 2021 00:00:00 +0000

Embarking on my journey into the realm of breast cancer detection has been both challenging and enlightening. In my pursuit of advancing the field, I am particularly drawn to the development of self-supervised and unsupervised approaches for breast cancer detection.

The application of self-supervised learning techniques allows for using the inherent information within the data itself, minimizing the reliance on labeled datasets and opening up new possibilities for more efficient and generalizable detection models. Simultaneously, the exploration of unsupervised approaches promises to uncover hidden patterns and structures within the breast data, potentially revealing novel insights that could contribute to the improvement of early detection methods.

As I navigate through this intricate landscape of research, my goal is to contribute to the ongoing efforts in the fight against breast cancer by innovating and refining detection methodologies. The intersection of cutting-edge technology, medical imaging, and artificial intelligence holds the promise of enhancing our ability to detect and combat this prevalent and impactful disease. In the pursuit of accurate, timely, and accessible breast cancer detection, my work strives to push the boundaries of what is currently achievable, with the ultimate aim of making a meaningful impact on healthcare outcomes.