To address the false alarm and missed detection problems caused by large differences in ship types, ships side by side and complex backgrounds in optical remote sensing images, a High-Efficiency Channel Attention Network (HECAN) model is proposed in this paper. We enhance the detection capability of the model for small-scale ships based on the EfficientDet model by using an efficient channel attention optimized feature extraction network to solve the channel information loss problem. Meanwhile, the Soft-NMS is used to improve the non-maximum suppression algorithm of the model to improve the recall rate of side-by-side ships. The results are compared with other object detection models and show that the HECAN model detects better on small-scale and side-by-side ships with an average precision of 92.46% and and a recall rate of 4.75% higher than the original EfficientDet.
Optical Remote Sensing Image Ship Detection Method for Improving EfficientDet
2021-11-12
571133 byte
Conference paper
Electronic Resource
English
Track Surface Defect Detection Based on EfficientDet
Springer Verlag | 2022
|Object Detection Using Vision Transformed EfficientDet
IEEE | 2023
|Track Surface Defect Detection Based on EfficientDet
British Library Conference Proceedings | 2022
|Track Surface Defect Detection Based on EfficientDet
TIBKAT | 2022
|Object Detection in Autonomous Maritime Vehicles: Comparison Between YOLO V8 and EfficientDet
Springer Verlag | 2023
|