Published Paper Details:

Automated image annotation plays a critical role in various fields such as autonomous vehicles, healthcare, e-commerce, and surveillance. Existing solutions often struggle with challenges like detecting small objects, handling occlusions, and providing fine-grained segmentation. To address these, this project introduces the Enhanced Deep ResNet50-SLT Model, which integrates multi-scale feature extraction, a contextual attention mechanism, and fine-grained semantic segmentation. The system is designed to detect and annotate objects with high accuracy, even in complex visual environments. Evaluation on datasets like MS COCO and PASCAL VOC demonstrates that the proposed model outperforms traditional methods with a mean Intersection over Union (IoU) of 80%, an F1-Score of 85%, and a 93% overall accuracy. This report covers the theoretical foundations, implementation methodology, and experimental results of the proposed system, highlighting its potential for real-world applications. While effective for simpler tasks, these methods struggled to capture spatial dependencies and contextual relationships, limiting their performance on complex, large-scale imagery. With the advent of deep learning, Convolutional Neural Networks (CNNs) became the dominant approach for remote sensing image classification and segmentation. CNNs are particularly adept at capturing local spatial features, making them well-suited for image classification and segmentation tasks. Several CNN-based frameworks have been developed for remote sensing, such as U-Net for semantic segmentation and Fully Convolutional Networks (FCNs) for pixel-wise classification.