Multilevel Fine-Grained Features-Based General Framework for Object Detection.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Existing object detection methods struggle with feature similarity, object size variability, and prediction inconsistencies in real-world scenarios.
• These limitations hinder the practical application of current object detection technologies.

Purpose of the Study:

• To propose a practical and generalizable object detector, the Feature Extraction-Fusion-Prediction Network (FEFP-Net).
• To address key challenges in object detection, including feature similarity, object size variability, and prediction accuracy.

Main Methods:

• Developed an adaptive fine-grained feature extraction network to capture details and avoid misclassification.
• Designed a bidirectional neighbor connection network for aggregating multilevel features to handle objects of different sizes.
• Implemented a task-specific prediction network leveraging spatial and channel information for enhanced localization and classification.

Main Results:

• FEFP-Net achieved competitive results on the MS-COCO dataset compared to state-of-the-art methods.
• Demonstrated satisfactory accuracy and generalizability in diverse fields like medical imaging, industry, agriculture, transportation, and remote sensing.

Conclusions:

• FEFP-Net effectively alleviates common object detection challenges.
• The proposed network offers a robust and generalizable solution for real-world object detection applications.