This study introduces an advanced model for deep semantic segmentation, targeting both natural and medical images. The model combines a Convolutional Neural Network (CNN) with Holistically-Nested Edge Detection (HED) and Spatial Pyramid Pooling (SPP) to improve accuracy in identifying and classifying image details. While CNNs are effective at detecting object shapes, they often overlook edges, which can lead to inaccuracies in applications like autonomous driving and medical diagnostics. By incorporating HED, the model ensures more robust edge detection, making it better suited for complex scenarios.
SPP is added to the architecture to enhance the model's ability to detect features across various image scales, capturing both large and small details effectively. The research tested this model using the CityScapes dataset, achieving high accuracy rates: 77.51% for small objects and 89.95% for larger objects. These improvements make the model especially valuable for tasks that require precise segmentation, such as identifying road features in real-time driving or detecting abnormalities in medical scans.
Overall, the proposed model offers a significant upgrade over traditional approaches, balancing high precision with computational efficiency. It opens new possibilities for real-world applications where accurate image analysis is crucial.
DOI: https://dx.doi.org/10.61927/igmin125
Full Text: https://www.igminresearch.com/articles/html/igmin125
Comments
Post a Comment